Download as pdf or txt
Download as pdf or txt
You are on page 1of 17

Defence Technology xxx (xxxx) xxx

Contents lists available at ScienceDirect

Defence Technology
journal homepage: www.keaipublishing.com/en/journals/defence-technology

AI-based small arms firing skill evaluation system in the military


domain
Rezoanul Hafiz Chandan, Nusrat Sharmin*, Muhaimin Bin Munir, Abdur Razzak,
Tanvir Ahamad Naim, Tasneem Mubashshira, Mokhlesur Rahman
Department of Computer Science and Engineering, Military Institute of Science and Technology, Dhaka, Bangladesh

a r t i c l e i n f o a b s t r a c t

Article history: Evaluation system of small arms firing has an important effect in the context of military domain. A
Received 25 July 2022 partially automated evaluation system has been conducted and performed at the ground level. Auto-
Received in revised form mation of such system with the inclusion of artificial intelligence is a much required process. This papers
12 February 2023
puts focus on designing and developing an AI-based small arms firing evaluation systems in the context
Accepted 24 February 2023
Available online xxx
of military environment. Initially image processing techniques are used to calculate the target firing
score. Additionally, firing errors during the shooting have also been detected using a machine learning
algorithm. However, consistency in firing requires an abundance of practice and updated analysis of the
Keywords:
Machine learning
previous results. Accuracy and precision are the basic requirements of a good shooter. To test the
Processing shooting skill of combatants, firing practices are held by the military personnel at frequent intervals that
Firing system include 'grouping' and 'shoot to hit' scores. Shortage of skilled personnel and lack of personal interest
Artificial intelligence leads to an inefficient evaluation of the firing standard of a firer. This paper introduces a system that will
automatically be able to fetch the target data and evaluate the standard based on the fuzzy systems.
Moreover it will be able to predict the shooter performance based on linear regression techniques.
Thereby, it compares with recognized patterns to analyze the individual expertise and suggest im-
provements based on previous values. The paper is developed on a Small Arms Firing Skill Evaluation
System, which makes the whole process of firing and target evaluation faster with better accuracy. The
experiment has been conducted on real-time scenarios considering the military field and shows a
promising result to evaluate the system automatically.
© 2023 China Ordnance Society. Publishing services by Elsevier B.V. on behalf of KeAi Communications
Co. Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/
licenses/by-nc-nd/4.0/).

1. Introduction weapon training institutions have their own training modalities to


conduct training on firing, evaluation and record keeping. The
Efficiency in small arms firing is the most fundamental skill for modalities followed during the training is mostly guided by human
an individual serving in military [1]. Weapon training, specially trainer and coaches who evaluate the errors and mistakes
training on small arms handling encompasses a major part of committed by the firer. They also guide and motivate them to
training curricula for newly inducted officers and men in military. become a good shooter following typical approaches followed in
Because small arms like rifles, pistols or submachine guns are military. The use of Artificial Intelligence can be a very useful tool
portable firearms which are treated as individual weapon. Armed for evaluating firing skill and analyzing shortcomings of the firer.
forces of any country plans and conduct training in such a way that Thereby it will enable tracking sequential skill development of a
each individual serving in military earns mastery on his personal firer which will ease up the professional responsibilities of a trainer
weapon. Coaches and trainers teach and guide the new inductees or coach. To sum up the whole military firing domain can be
so that they become skilled shooter in course of time. Different divided into two parts: (i) Firing Evaluation System (FES) and (ii) To
analysis the performance of the firer. FES consists of target scoring
and error in grouping part and the performance analysis part,
* Corresponding author. which can be performance measure and performance prediction.
E-mail address: nusrat@cse.mist.ac.bd (N. Sharmin). To simplify the whole concept in this paper, we have divided the
Peer review under responsibility of China Ordnance Society

https://doi.org/10.1016/j.dt.2023.02.024
2214-9147/© 2023 China Ordnance Society. Publishing services by Elsevier B.V. on behalf of KeAi Communications Co. Ltd. This is an open access article under the CC BY-NC-
ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Please cite this article as: R.H. Chandan, N. Sharmin, M.B. Munir et al., AI-based small arms firing skill evaluation system in the military domain,
Defence Technology, https://doi.org/10.1016/j.dt.2023.02.024
R.H. Chandan, N. Sharmin, M.B. Munir et al. Defence Technology xxx (xxxx) xxx

discussion based on this two main part, FES and the firer perfor- attempts to provide a brief overview of the military firing system to
mance analysis (see Figs. 5 and 6). assimilate the readers with the process. Section 2 provides a review
Traditionally training on firing in military institutions follows its on related works and Section 3 describes in detail the methodology
own typical methods and drills. The evaluation and record keeping followed to implement the idea of the project. Section 5 shows the
methods are partially or entirely manual [1]. A package of drills is results of the proposed system as per the prototype formulated.
followed in the firing range for smooth conduct of the firing and to Section 6 enumerates the evaluation data of the prototype based on
ensure safety and security of all. As soon as firing of a group of user feedback studies. Section 7 will have the concluding remarks
soldiers is completed, they along with the evaluators and coach on the whole paper.
goes for target checking. Marks and points are calculated manually
by counting total number of shots on the target and by measuring 2. Military firing system
distance among bullet impact points using a measurement scale.
Thereafter, the result of each firer is manually recorded to a data- Skill in firing is always considered as an important trade of a
base or in registers. Here the difficulties arise because, though the good soldier. An organized and well-structured training package is
earned points of each firer is recorded but the pattern of shoots followed to develop the firing skill of a soldier during his induction
cannot be recorded which is a vital information for firer's rectifi- training in the military. Even after his induction in the military unit,
cation of error and further development. Firing pattern assists a he needs to go through regular training to upkeep his quality as a
firer in comprehending his own mistakes as well as acts as a tool to good firer. The firing training includes delivery of the knowledge
fix them. By continuous analysis of the firing pattern, a person can about the weapon and requirement of a good shot. Not only
minimize his mistakes and errors during firing and can takes knowledge but also a good built-up of hand muscle and capability
deliberate measures to overcome the issues. In the existing process of following firing cycle efficiently is a must to be a skilled firer. So,
of evaluation, the firer remains unaware of his past mistakes and condition of the weapon and capacity of the firer both affect the
lose focus to improve his firing efficiency. Again, the existing precision and accuracy of the firing results.
evaluation and record keeping process is time consuming, effort Under ideal condition, if a weapon is fired with a correct aim, the
centric and less efficient. Thus, a computer aided automated eval- bullet should hit the bull on the target. But this does not happen
uation and record keeping process will be useful for firing skill due to two basic reasons: either the weapon is not zeroed or the
development of the military personnel. fault remains with the firer. If a weapon is not properly zeroed, then
There is a vast use of image processing techniques and machine a good firer may bring accurate firing on the target which may not
learning in sports domain [2e7]. In literature, the performance of have precision. A weapon is zeroed by conducting grouping fire on
the sportsmen in various sports have been predicted through ma- the target by a good firer. A group is the collective pattern of the
chine learning. Even in different firing competitions image pro- shots fired aiming each time at a single point, without any major
cessing techniques are widely used to calculate shooting scores. changes in firing position. The firer gets the chance of shooting five
There are examples of performances measurement and prediction bullets at the target from a 100-m distance in one shooting session.
for sportsmen with certain factors as input. In the Olympics, the The size of the group is a measure of a soldier's skill in weapon
biathlon performance [8] is predicted using machine learning and firing. The smaller the size of the group the more skilled the soldier
related algorithms. The machine is initially fed with previous years' in firing.
firing data. Thereafter the factors impacting shooting performance
are studied using exploratory data analysis. This assists in various
machine learning models for predicting future hits and misses. But 2.1. Firing evaluation system
considering the available open sources information it can be said
that machine learning has not been used yet to measure or predict The firing evaluation system depends on the target scoring and
firing performance of a soldier. errors in the grouping. Hence We divided this part of the discussion
The performance measure is therefore of utmost importance to into two parts: (i) Target scoring system and (ii) errors in grouping.
guide an evaluator in understanding the firing pattern of a firer.
Side by side, it assists a firer in comprehending his own mistakes as 2.1.1. Target scoring system
well as acts as a tool to fix them. It carefully compares the individual The maximum distance between the two furthest impacted
attributes within a group and works as a key element to stan- bullets on the target is the size of the group. If the group size is
dardize the performance. By continuous analysis of the firing more than 10 inches, then the fire is declared as wash-out and he is
pattern, a person minimizes the mistakes and errors he/she can considered to be exposed to errors in following the firing cycle and
commit in the process and takes deliberate measures to overcome poor weapon handling. The pattern of impacted bullets on the
the issues. To our knowledge, no specific literature study based on target of a standard firing is shown in Fig. 1. If the group size is 1e4
the performance of the military using machine learning techniques inches the firer is classified as marksman (Fig. 1(a)), if the group size
has been found. is 5e7 inches, they are classified as first-class firer (Fig. 1(b)) and if
This paper will mostly emphasize automating the current firing the group size is 8e10 inch, they are called Standard firer (Fig. 1(c)).
evaluation system and contribute to the following: The grouping of fire has a direct relation with the zeroing of the
weapons. A perfectly zeroed weapon creates a marksman shot close
(1) Development of an automatic target scoring system as well to the center of the target (also called "bull"). A non-zeroed weapon
as various error detection system using Image Processing and will formulate the same group away from the bull. The error shots
machine learning techniques from a real-time data-set. are caused mostly due to the non-zeroed weapons apart from the
(2) Proposal for a concise measurement of firer performance mistakes made by a firer.
with comprehensive relation among various performance The grouping of fire has a direct relation with the zeroing of the
criteria using the fuzzy logic system. weapons. A perfectly zeroed weapon creates a marksman shot
(3) Analysis of an individual's firing performance and predict the closure to the center of the target (also called “bull”). A non-zeroed
future outcomes using the linear regression model. weapon will formulate the same group away from the bull. The
error shots are caused mostly due to the non-zeroed weapons apart
We divided the rest of the paper in to seven sections. Section 1 from the mistakes undertaken by a firer.
2
R.H. Chandan, N. Sharmin, M.B. Munir et al. Defence Technology xxx (xxxx) xxx

Fig. 1. (a) Marksman; (b) First class firer; (c) Standard firer.

2.1.2. Errors on grouping firer on the point of aim or on the front sight tip. Such a group is also
A firer has to learn and follow some basic principles of firing like created if the firer fails to make the correct trigger pull. If the firer
steady hold, correct aim, and correct trigger control. A firer has to has a number of errors and fails to make the correct hold while
ensure the correctness of certain aspects while he fires, like correct firing, then the impacted bullets do not create any pattern on the
body alignment with the target, correct sight picture, breathing target. These errors are shown with relevant figures in Table 1 for
control, correct trigger press, and follow through. Again, he has to further assimilation with the errors.
abide by a firing cycle to bring all the shots at the point of aim at the
target. The weakness of a firer to follow some firing principles is
clearly projected on the target. An experienced firing trainer can 2.2. Performance analysis of firer
easily identify the mistakes of a firer by seeing the pattern of the
group created on the target and can guide the firer to correct the Methods and techniques to measure the firing efficiency of
mistakes in the subsequent firing practice. military personnel are different in countries around the world.
There is four defined pattern of bullets'impact on the target Again, based on chronological development and progressive
which represents four common errors committed by a firer. These improvement of firing skills, military personnel can be grouped into
errors are shown with relevant figures in Table 1. Long horizontal three categories: trainees, new soldiers, and trained soldiers. In
error is caused by horizontal movement of the left hand due to for course of time, as the soldiers develop experience and skill in firing
poor holding. Long vertical error caused by wrong breathing con- they are trained in advanced firing techniques and are encouraged
trol or due to error in alignment of back sight U and front sight tip of to fire in varying situations with varying ranges. To give them
the weapon. A bifocal group is created due to excessive stress of the battlefield experience they are trained to fire as a team and adapted
through battlefield inoculation training. But, whatever is the state,
skill, and group of the firer they are basically evaluated based on
Table 1 two types of firing namely grouping fire and shooting to hit fire.
Different errors. Both these two types of firings are conducted at different distances
Scrial Error type Figure Remarks where the individual fires his personal weapon being at different
firing positions, e.g. laying, sitting, kneeling, standing position, etc.
1 Long horizontal error Recognized
Grouping fire is basic teaching for every soldier and is the main tool
for judging firing skills. Grouping fire is usually conducted at a 100-
m distance on a 3-feet by 3-feet square-shaped target. This type of
firing is usually conducted in a lying position. The diameter of a
2 Long Vertical error Recognized circle containing all five shots of the firer is known as his grouping
capacity. Size of the groups for counting points varies basing on the
type of weapon used in firing. For rifle firing, if the firer is able to
keep all five shots within 4 inches on the target, he gets 5 points.
3 Biforcal error Recognized For a group size of 5e7 inch and 8e10 inch 4 and 3-points are
allotted respectively. More than 10-inch group is considered as
wash-out or fail score. Shoot-to-hit fire is conducted at different
distances starting at 300, 200, 100 and ending at 25-m distance.
Soldiers fire their personal weapon being at different positions
4 Scattered group Recognized
from different distances and proceed nearer to the target. Targets
on which the firer take shots are also different in size and shape for
different distances. In case of shoot to hit fire, the firer gets one
point for each shoot hitting the target. So, at the end total points are
5 Impatient shot Unrecognized aggregates of points in grouping fire and shoot to hit fire. Rules of
counting of points to measure performance is same for male and
female soldiers irrespective of their age and service length. How-
ever, to encourage and recognize the performance of trainee sol-
diers or recruit, new soldiers joining the units and female soldiers,
3
R.H. Chandan, N. Sharmin, M.B. Munir et al. Defence Technology xxx (xxxx) xxx

they are categorically judged within their peers and best per- from one another. The primary objective of the firer, in accordance
forming among them are rewarded. with the study on the causes of delivery mistake conducted by
Storhm et al. [17], is to reduce the time between rounds so that the
3. Related works shot distribution approaches the MPI and maintains the MPI
around the bullseye. The performance metric for the firer is ulti-
Researchers have previously done some works related to the mately influenced by these shot distributions.
proposed firing skill evaluation system. This section will focus on Another important parameter precision is defined as the value
the previous research works on three main parts: (i) Automatic of the precision parameters, i.e., stronger precision qualities are
target score through image processing; (ii) Performance measure represented by smaller hit pattern dimension sizes and lower
through fuzzy system; (iii) Performance prediction using machine precision measure values. The average hit is one of the unique as-
learning. pects of the hit pattern, and it may be computed or visually
determined. However, in the case of firing, the term accuracy refers
3.1. Firing evaluation system: automatic target score through image to the relationship between the hit pattern features and the di-
processing mensions and attributes of the specified target. Zivkovic et al. [18]
designed a non-material high-resolution target acquisition system
Ali et al. [9] proposed a computer vision based automatic scoring using IR sensor. They controlled the hardware system with a simple
system for shooting targets using simple image processing tech- algorithm to determine the precision and accuracy of the firing
niques like morphological processing and hysteresis thresholding. system.
Their proposed algorithm achieved 98.3% accuracy tested on 100 For determining these performance metrics more accurately,
images of different number of bullet hits. Another noted research fuzzy logic is being popular day by day. A fuzzy logic system (FLS)
done by Ding et al. [10] proposed a machine vision based scoring [19] is a special mechanism that can manage numerical data and
system using the back lighting illumination system. Their main linguistic information at the same time. It's a nonlinear mapping of
success was that, the system could recognize the overlapped holes an input data (feature) vector to a scalar output or converting
on real time and therefore, could perform quickly with high accu- numbers to numbers which is done through three steps: Fuzzifi-
racy. Ye et al. [11] also developed an autocounting system using cation, Inferencing, and Defuzzification [19].
some selected techniques of image processing and image recogni- Research Gap. There are cases where performance characteris-
tion. However, with the growing technology of the pattern recog- tics are used for shooting, but nowhere are they used in a way that
nition ellipse fitting algorithm, some researchers have worked with specifically measures total performance. A fuzzy inference method
advanced techniques for instance, Soetedjo et al. [12] implemented is never employed to evaluate the performances.
an embedded camera system and used homography transform.
Their system could perform accurately regardless of camera angle 3.3. Performance prediction using machine learning
and viewpoint. Kusuma et al. [13] have used several image pro-
cessing techniques like target ring detection, and perspective Performance analysis helps to determine the level of expertise
transform with other traditional techniques. They achieved an ac- in individual firing. In the case of bullet shooting or firing, the
curacy of 91%. shooting device/weapon handling reflects a greater degree of
A firer is prone to errors, the majority of the time, which is due overall performance. The gun handling aspects of firing are hold,
to a lack of understanding of how to fix errors. blackIf a firer un- stability, aiming accuracy, triggering cleanness, and time taken on
derstands the causes of his errors, he is likely to overcome his target. Ihalainen et al. [20] conducted a study to discover the most
limitations [14]. Firers’ mistakes and defective weaponry are the significant factors to determine the performance in elite-level air
two most common causes of errors. Hatamleh et al. designed a low- rifle shooting technique. They came up with four important pa-
cost system for visual laser shooting practice system where they rameters using regression analysis and Principal component anal-
tried to overcome these errors using a developed image processing ysis found two more factors.
algorithm [15]. SVM is a classical supervised learning model that can analyze
There are several sources of dispersion that lead to firer's faults data and recognize patterns. It is based on statistical learning the-
and worsen firing performance in terms of hit probability. blackIssa ory and is widely used to summarize and review research and
et al. addressed some of these sources of dispersion but didn't development works. It can generate highly accurate results for the
develop any separate mechanism to overcome these issues [16]. diagnosis and finding errors. SVM is basically a collection of su-
The cause of faults leading to inaccuracy and dispersion must be pervised learning techniques for classification and regression [21].
recognized to improve firing effectiveness. Ammunition dispersion They are members of the generalized linear classification family.
error, gun dispersion, aerodynamic jump, and aiming error are only SVM has a unique virtue in that it minimizes the empirical classi-
a few of them which were addressed by Seotedjo et al. in their fication error while also increasing the geometric margin. As a
research [12]. There's also vertical error, horizontal error, bifocal result, SVM is also known as Maximum Margin classifier. SVM maps
error, scattered error, and so on. the input vector to a higher-dimensional space in which a maximal
Research Gap. The majority of the aforementioned works were separating hyperplane is built. It allows Structural Risk Manage-
created in a simulated setting rather than a real-world military ment (SRM) meaning it can tune the capacity of the classifier to the
setting. Additionally, no work has used machine learning to available amount of training data. Thus, SVM can prevent
recognize the mistake grouping. phenomenal overfitting by balancing the complexity of models
with their fit to the data. Though there are examples of using SVM
3.2. Firer performance analysis: performance measure through to find out employee performances [22] in an organization, there
fuzzy system are no instances of using this system to evaluate firing
performances.
The average impact position of a shot dispersion is referred to as Regression is a very popular and basic tool for prediction anal-
the mean point of impact (MPI). A shot distribution that places the ysis. Regression seeks to answer two questions. First, can an
MPI closer to the target's bullseye aids in high accuracy. On the outcome (dependent) variable be predicted using a set of predictor
other hand, accuracy suffers if the rounds are placed too far apart variables? Second, based on the size and sign of the beta estimates,
4
R.H. Chandan, N. Sharmin, M.B. Munir et al. Defence Technology xxx (xxxx) xxx

what factors, in particular, are significant predictors of the outcome


variable and how do they affect the outcome variable are discussed.
Almost every sector of real-world applications has an impact with
the solution of regression problems [23].

4. Methodology

The proposed method is carried out in two separate phases in


gradual stages. The first phase consists of the generation of scores
and detection of various types of errors using machine learning and
various image processing algorithms based on real-time images of
firing. The second phase consists of performance analysis through
Fuzzy Inference System and the linear Regression technique is
utilized to predict the future performance probabilities of a fire.
The Whole methodology part has been divided into two parts: Fig. 2. Coordinates of bullet holes.

 Phase 1: Firing evaluation system-generation of individual


scores and detection of errors. 4.2. Phase 2: performance analysis of firer
 Phase 2: Performance analysis of firer has also been divided into
two parts: (i) Performance analysis using fuzzy logic and (ii) 4.2.1. Performance measure using fuzzy logic
performance prediction using linear regression. Fuzzy logic helps to find out a point between any two real
numbers. Most of the cases, the performance measure of a person
or a system is obtained from a relation between few variables that
4.1. Phase 1: firing evaluation system-generation of individual constitutes some real numbers. As a result, fuzzy logic can be
scores and detection of errors applied to find out the performances of a firer based on some pre-
selected variables that influences the firing performance measures.
The Structural Similarity Index (SSI) algorithm is used to sort the
bullet holes from the target that requires two images a reference 4.2.1.1. Problem formulation. Practically, it is difficult to determine
image and a target image from the same image capture. It then the parameters that cause a change in firing performances. From
measures picture quality degradation as a result of processing such the studies, it is observed that firing performances are connected to
as data compression or data transmission losses. practice and past experiences. However, only practice and previous
The faults made by a firer lead to various firing errors and these experiences cannot be considered as the core performance measure
errors can be divided into 5 types. The common errors in the case of in general. The number of firings conducted per single day and
grouping fire are Long Horizontal Error EHor, Long Vertical Error period of the interval from the previous firing are also accountable.
EVer, Bi-focal Error EBF, Scattered Group ESc and Impatience Error The effect of the season cannot be overlooked too. Therefore, a
EIm. performance measure for firing includes a wider range of param-
The pattern of these errors can be calculated using 2 approaches. eters. For convenience, the parameters selected for analyzing the
The first approach is to detect the error analyzing the elliptical performance are group, number of firings, interval, season, service
pattern. An ellipse is drawn calculating the variance of the bullet length and average group.
holes and taking the mean value as the center. Then the properties
of the ellipse was analyzed to find the particular error type.
4.2.1.2. Solution. The firing performance of an individual is
measured with the help of Fuzzy Logic System (FLS) using neces-
sary membership functions (MF). The MFs are selected in accor-
Error Detection by Ellipse Estimation dance with the variables that provide satisfactory results.
Input: Array of points, P ¼ [P0,P1,P2,P3] P4 where Pi ¼ (xi,yi).
Output: corresponding error E.
Step 1 / Find array of distances between points.
4.2.1.3. Fuzzy Logic System. A FLS is a system that performs its
Step 2 / Segregate x and y variables from points array. calculations based on precise knowledge which is stated by lin-
Step 3 / Calculate variances of x and y as xvar and yvar. guistic variables. The approach of FLS imitates the way of decision
Step 4 / Find Ecenter ¼ (mean.x,mean.y). making in humans that involves the intermediate possibilities be-
pffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffi
if xvar > 3 yvar
tween YES and NO. The conventional logic block that a computer
then E ¼ EVer
pffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffi can understand produces a definite output as TRUE or FALSE, which
else if yvar > 3 xvar
then E ¼ EHor can be interpreted as human YES or NO. Unlike machines, the hu-
P
else if (max(D))2 > ni¼0 ðDnewðiÞ Þ2 ,where, Dnew ¼ Dmax(D) man decision making includes a range of other possibilities be-
if max(D) ¼ ¼ Dn tween YES and NO, such as CERTAINLY YES, POSSIBLY YES, CANNOT
then E ¼ EIm SAY, POSSIBLY NO and.
else E ¼ EBF
4.2.1.3.1. CERTAINLY NO. The FLS works with the same levels of
else E ¼ ESc
possibilities of input to get the definite output. The system mainly
comprises three basic processes:
The distinctive features assumed for the points to fulfill the
(1) Fuzzification (input stage) to receive input values and
criteria for errors are the pairwise Euclidean distances between
translate them to Fuzzy type inputs via transfer functions
observations in n-dimensional space (Fig. 2). A supervised learning
such as triangular, trapezoidal, Gaussian distribution curve,
technique is used to predict the error class. The training data and
sigmoidal functions, etc. A fuzzification unit supports the
test data were split into 80e20 ratio to achieve good accuracy.
application of the transfer functions based on the suitability
5
R.H. Chandan, N. Sharmin, M.B. Munir et al. Defence Technology xxx (xxxx) xxx

of the function in relation to inputs, and converts the crisp of employees and organizations. The same idea can be applied in
inputs into similar fuzzy inputs. the measurement of firing performance provided exact inputs are
(2) Fuzzy Inferencing (processing stage) for generating inferred fed to the system.
results based on previously fuzzified values and linguistic IF/
THEN statements. two important methods of FIS, having
4.2.1.4. Membership functions. A membership function (MF) is an
different consequent of fuzzy rules are the Mamdani Fuzzy
important part of fuzzy logic calculations that provides information
Inference System (MFIS) and the Takagi-Sugeno Fuzzy Model
that maps the inputs onto the linguistic variables. An MF is mainly a
System (TSFMS). curve that defines how each point in the input space is mapped to a
(3) Defuzzification (output stage) is applied for totaling the
membership value. The MF has a range within which the members
result using linguistic rules and converting those into output
are defined. The parameters in an MF are defined to relate to the
values. It is realized by a decision-making algorithm that
presumable results and can also be named as variables. Various
selects the best value based on a fuzzy set.
transfer functions such as triangular, trapezoidal, Gaussian distri-
bution curve, sigmoidal functions etcetera are used as MFs with
The FLS can be implemented in systems with various sizes and
crisscrossed parameters for fuzzification of the input variables. The
capabilities ranging from small micro-controllers to large control
triangular MFs are mostly used in the performance measurement
systems. Nowadays, it has become a core tool in the artificial in-
as it provides the concrete result. Different Membership Function
telligence being experimentally used in performance measurement
used as Input variables are as discussed below.

Fig. 3. The Membership Functions with its details.

6
R.H. Chandan, N. Sharmin, M.B. Munir et al. Defence Technology xxx (xxxx) xxx

(1) Groups: The shot grouping/group has a direct relation with minimum two grouping fires are conducted, therefore its
the performance. The better is the grouping performance, range is taken slightly less than the ongoing group. The range
the more skilled is the firer. The range is taken from 0 inch to is determined as between 0 and 36 inches.
53.5 inches based on the total target size and relevance with
best and worst cases The MFs used in the input variables with its details about ranges
(2) No of firings: It is tiresome to fire continuously throughout and types are shown in Fig. 3.
the day. Therefore, the number of firings per single day might A graphical representation of the MFs used as the FLS inputs are
decline the performance. The range is assumed to be from no shown below to further assimilating the idea of various types of
firings to 18 times of firing events for an individual per single functions.
day.
(3) Interval: There is a relation with the firing performance and
4.2.1.5. Fuzzy rules. Fuzzy Inferencing System at the processing
the continuation from the previous firing. The interval be-
stage works with the rules that direct the input variables towards a
tween two firings may affect the performance. Its range is
generalized output. The rule base of the proposed fuzzy perfor-
taken between 0 and 90 days.
mance analysis model is defined in three steps; firstly, three inputs
(4) Season: The seasons have a great deal of influence on the
about their performance provides performance measure 1; sec-
firing performance. The visibility an daylight varies in
ondly, the first and other two inputs give performance measure 2
different seasons. Therefore, six seasons are also taken as
and finally, the first and last input maps into performance measure
input to relate with the performance.
3. The numbers of rules were computed based on the permutation
(5) Service length: As a man in uniform ages, his/her experience
with a number of membership functions and numbers of perfor-
about techniques of fire grows as well. The service length is
mance measure parameters. The numbers of factors are nine and
therefore the sixth input in the FIS. The range of service
the number of membership functions that provide performance
length is considered from 0 to 36 years.
measures is three. The total number of rules, therefore, became
(6) Average group: To generate the performance measure, the
93 ¼ 27. The rules are shown in Tables 2e4 respectively (see
average group size of an individual is also of overwhelming
Fig. 7).
importance. As the average group is the measure when

Fig. 4. Input: No of Firings using trapezoidal MFs.

Fig. 5. Input: Interval using gaussian MFs.

7
R.H. Chandan, N. Sharmin, M.B. Munir et al. Defence Technology xxx (xxxx) xxx

Fig. 6. Output: Performance 1 using triangular MFs.

4.2.1.6. Defuzzification and results. The three performance mea-


Table 2
Fuzzy rules for performance 1.
sures (see Figs. 4e7) obtained from the initial fuzzy model are used
as inputs for the second fuzzy model. Thereafter certain new rules
R/N If Then
lead to a new calculation step to finally define the performance of
Group No_of_firings Interval Performance the firing. Here, the centroid defuzzification is used the find the
1 49e54 inch Or 1e2 Or 00e10 Minimal crisp performance value. The following Fig. 10 shows an instance of
2 43e48 inch 3e4 11e20 Dismal defuzzification module in this FLS.
3 37e42_inch 5e6 21e30 Poor The total model is used in a MATLAB Simulink model and pro-
4 31e36_inch 7e8 31e40 Mediocre
vides the required analysis of performance as a concrete value.
5 25e30_inch 9e10 41e50 Average
6 19e24 inch 11e12 51e60 Decent Fig. 11 provides an example of measurement of performance using
7 13e18 inch 13e14 61e70 Good the FLS in total.
8 7e12 inch 15e16 71e80 Excellent
9 1e6 inch 17e18 81e90 Extraordinary
4.2.2. Performance analysis using linear regression
Performance analysis is required to enhance the mastering
process in a particular field. One of the tools used to conduct the
Table 3 performance analysis in AI is the Linear Regression model. Linear
Fuzzy rules for performance 2. regression helps to identify the best matches for the data points on
Group Season Svc_length/yrc Performance_2 a linear plot. As a result, it can be used to forecast output values for
10 49e54 inch Or Rainy Or 36 Minimal
inputs that aren't included in the data set, with the assumption that
11 43e48 inch Rainy 4 Dismal those outputs will fall on the line.
12 37e42_inch Winter 32 Poor
13 31e36_inch Late_autumn 8 Mediocre 4.2.2.1. Dataset. Practically, it is difficult to determine the future
14 25e30_inch Autumn 28 Average
15 19e24 inch Summer 9 Decent
performance of a firer. But as mentioned earlier, there might be
16 13e18 inch Summer 18 Good some connection between future firing performances and the
17 7e12 inch Spring 16 Excellent number of practices, past performances, and some other factors. For
18 1e6 inch Spring 20 Extraordinary convenience, the parameters selected for analyzing the perfor-
mance are the number of firings conducted per single day, the in-
terval between two firing sessions, seasonal effects, length of
Table 4 service of the firer, and their average group. These are the predictor
Fuzzy rules for performance 3.
data of the system. The target data are the performances obtained
Group Avg_gp Performance 3 from the FLS corresponding to the mentioned values. It's worth
19 49e54 inch Or 36 inch Minimal mentioning here that due to some constraints for real-time data
20 43e48 inch 32 inch Dismal collection from the military domain, the size of dataset is limited.
21 37e42_inch 28_inch Poor
22 31e36_inch 24 inch Mediocre
4.2.2.2. Training. The retrieved characteristics are used as inputs to
23 25e30_inch 20_inch Average
24 19e24 inch 16 inch Decent create a model from the data and make predictions. The dataset
25 13e18 inch 12_inch Good gathered provides two inputs to fit where X remains as the pre-
26 7e12_inch 8 inch Excellent dictor variable and Y acts as the target variable. As the cost function,
27 1e6 inch 4 inch Extraordinary the Mean Squared Error (MSE) function is used, which is the
average of squared error occurred between the predicted values
and actual values. The model performance or the goodness of fit is
A selection of surface graphs showing the relationship between checked through process of optimization. For easy understanding,
input and output variables are shown in Figs. 8 and 9 respectively. the R-squared method as shown below is used to apprehend the

8
R.H. Chandan, N. Sharmin, M.B. Munir et al. Defence Technology xxx (xxxx) xxx

Fig. 7. Output: Performance 2 using generalized bell MFs.

100 after pairing with the X (Predictor) data and Y (Target) data.

5. Experiments and results

In order to determine the accuracy of the system, the entire


experimental method was split into two major modules, namely
the Firing Evaluation System and Performance Analysis of Firer. The
goal was to comprehend the system's strengths and weaknesses
and to suggest the changes required to improve its stability. The
photographs used in this project came from a variety of military
formations and real-time units. To ascertain the frequency of
inaccuracies, the results of manual calculations compared to the
photographs were also gathered. In addition, the firing records of
Fig. 8. Interval and group vs. performance.
each soldier were discussed, and the necessary data sets were
created to conduct the trials.

5.1. Firing evaluation system

We divided the evaluation of the firing evaluation system into


three parts: (i) Bullet hits detection; (ii) Calculations of groups; (iii)
Error detection.

5.1.1. Bullet hits detection


The experiment was conducted to find out the success rate of
the detection of bullet holes.

5.1.1.1. Dataset. The distribution of the dataset for various error


types has been illustrated in Fig. 12.
Fig. 9. Average group and group vs. performance.

5.1.1.2. Experimental setup. At first, the images of fresh targets were


optimization level of the model. captured using a 6-megapixel phone camera at a distance of 25 m
from the targets with camera stands. Then the firing was done in
P  !
ðy:true  y:predictedÞ2 the standard manner i.e. five bullets grouping fire at the 100 m
R ¼ 1  P
2  (1) range. After the firing, the target Images were taken again in the
ðy:true  meanðy:trueÞÞ2 same manner from the same fixed distance. In this manner, 16 sets
of Images were taken from eight targets. Thereafter, the SSIM
module was used to find the similarity and detect the bullet holes.
The results are shown in Tables 5 and 6.
4.2.2.3. Prediction. The trained model can now predict future per-
formance based on input data, thanks to the relationship between 5.1.1.3. Result. From Tables 5 and 6, it is clearly understandable that
the datasets. The input data is adjusted to fit the elements of the the program can detect all the shots quite correctly. The tables have
dataset. The proportion of material that is most relevant is calcu- the following columns: Serial, Images after Firing, Bullet Hits
lated and delivered as an output. The associated relation indicates Detected (Indicated by Red), SSIM Scores (percentage), and Accu-
the possible percentage of performance at a scale between 0 and racy (percentage). The images from serial numbers 2 and 4

9
R.H. Chandan, N. Sharmin, M.B. Munir et al. Defence Technology xxx (xxxx) xxx

Fig. 10. Centroid defuzzification method.

Fig. 11. Simulink model for Defuzzification module.

demonstrate that the system correctly identified a hit from two quality, the way the bullet hit, and the brightness variations be-
gunshots as a single hit. But after morphological processing, these tween the images.
are once again distinguishable. The technique yields similar results
while using freshly printed target paper as well as frequently used 5.1.2. Calculation of groups
target paper. The fact is that the bullet shots can be determined The images found from the SSIM module are further processed
with 100% precision is a success. The proportion of similarity be- and subjected to group calculation module. The groups are only
tween the photos taken before and after firing is determined by the calculated if five bullet marks were found by the system. Otherwise,
SSIM score. The variations in SSIM scores are caused by the image the system declines to calculate the results.
10
R.H. Chandan, N. Sharmin, M.B. Munir et al. Defence Technology xxx (xxxx) xxx

Table 6
Accuracy of bullet hits detection.

Serial After firing Bullet hits detected SSIM score/% Accuracy/%

5 0.997 100

6 0.988 100

Fig. 12. Error distributions.

Table 5 7 0.997 100


Accuracy of bullet hits detection.

Serial After firing Bullet hits detected SSIM score/% Accuracy/%

1 0.993 100

8 0.998 100

2 0.986 100

trial, the groups determined by the system were larger than the
3 0.996 100 ones determined manually. Those were brought about by slight
flaws in the targets' geometrical design. A 0.2-inch allowance can
be added to the firers to encourage them and help alleviate the
problem.
A chart (Fig. 13) is prepared to understand the differences in
4 0.967 100 results more clearly. Here the various targets are listed along the X-
Axis and Y-Axis shows the measure of grouping in inches. The green
bars represent the results calculated manually and the orange bars
show the results found by the Automatic Firing Target Scoring
System.

5.1.3. Error detection


In the Error detection module, both the error detection by use of
5.1.2.1. Experimental setup. To collect the experimental data, a ellipse and error detection by support vector classification modules
7.62 mm Sub-Machine-Gun type 1956 was selected as the primary were tested. The results from the system were matched with actual
firing weapon. A total of 8 firers were detailed to conduct firing. data to check and analyze the accuracy of the modules. Experi-
They fired 3 times. Only one set of data was taken for experimental mental setup Different sets of target data were used, where the
calculations. system could easily identify the errors from analyzing the data. To
tally all types of errors, the images were manipulated to make
5.1.2.2. Result. The information gathered through the manual images of error shots. This was because manually it was difficult to
measurement method using a twelve-inch scale as well as through collect images that has all types of error-biased shots. The images
the suggested software is provided in the following Table 7. As a taken as the inputs are shown in Fig. 14.
result of the system's ability to calculate results in fractions, the
table displays promising states. In addition to calculating the shot 5.1.3.1. Result. Table 8 shows the errors detected by the system
deviation, the automated system also delivers results of shot using both modules. The data indicates that there are situations
grouping. As a result, it helps with figuring out the corrections where only one of the modules is able to identify the error correctly.
needed for weapon zeroing. In order to achieve these values, a Similar types of errors can have various firing patterns. Due to the
manual procedure would have taken much longer. The variation in limited amount of data, however, the machine learning component
measurement between the automatic system and the manual occasionally fails to detect the actual issue. Systems are occasion-
scoring system (A-B). is also displayed in table. A manual procedure ally created through error identification using a mathematical
is flawed because there may be little errors made when figuring out technique. It incorrectly conflates the impatient shot with any
the center-to-center distances between the bullet shots. Due to this, horizontal or vertical flaws. If both modules have the same error,
two firers may occasionally achieve comparable outcomes while only one error will be shown by the system. More than one error
having little variations in their shot grouping. The software is quite detected by the modules will cause both errors to be displayed.
useful in reducing this problem. In certain instances during the However, the success of the system is that it could identify the

11
R.H. Chandan, N. Sharmin, M.B. Munir et al. Defence Technology xxx (xxxx) xxx

Table 7
Grouping results: Manual and software-driven.

Target serial Manually calculated Result from automatic firing target scoring system Deviation in
group in inches (A) inches (A-B)
Group in inches (B) Classification (C) Vertical correction (D) Horizontal correction (E) Error (F)

Target No-1 12.5 12.6 Wash out 1.24 rotation downwards 0.51 mm towards left Bi-focal error 0.1
Target No-2 6 5.76 First class shot 2.79 rotation upwards 0.04 mm towards right Nil þ0.04
Target No-3 12.5 12.48 Wash out 1.24 rotation upwards 0.78 mm towards right Impatient error þ0.02
Target No-4 11 10.8 Wash out 0.15 rotation downwards 0.56 mm towards right Nil þ0.2
Target No-5 9 9 Standard shot 0.45 rotation upwards 1.62 mm towards right Nil 0
Target No-6 10 9.84 Standard shot 0.87 rotation downwards 0.10 mm towards right Nil þ0.16
Target No-7 25 25.2 Wash out 0.70 rotation upwards 0.22 mm towards left Horizontal error 0.2
Target No-8 4 4.25 Marksman shot 2.43 rotation upwards 0.03 mm towards left Nil 0.25

corresponding errors with at least one of the modules. More data


sets can be fed into the system to make it more robust, which will
improve accuracy.

5.2. Performance measure

To find the performance measure, the performances of an in-


dividual were fed to the system. The fuzzy system thereafter
calculated the performance measure on a scale of 100.

5.2.1. Experimental setup


The factors to calculate the performance were stored in a
database as numbers. These data were stored against a date to
comply with real-time scenarios. The data was then analyzed by the
Fig. 13. Grouping comparison chart.

Fig. 14. Error images.

12
R.H. Chandan, N. Sharmin, M.B. Munir et al. Defence Technology xxx (xxxx) xxx

Table 8
Error detection.

Serial Before firing Error by ellipse Error by SVM Actual Error Incorrect Reasons
module

1 Long horizontal Long vertical Long horizontal Error by SVM The pointwise distances for horizontal and vertical errors are
error error error quite similar

2 Bi-focal error Bi-focal error Bi-focal error e e

3 Scattered group Scattered group Scattered group e e

4 Impatient shot Long vertical Impatient shot Error by SVM More data set would provide accuracy in impatient shots
error

5 Bi-focal error Bi-focal error Bi-focal error e

6 Long horizontal Impatient shot Impatient shot Error by The serial wise distribution of points leads to calculations
error Ellopse

7 Impatient shot Impatient shot Impatient shot e e

8 Long vertical Long vertical Long vertical e e


error error error

MATLAB FIS module to calculate and store crisp results in a separate group. Because a slight change in the group gets the overall per-
database table. Two instances of the performance measures formance measure affected. However, the results are satisfactory as
calculated by the system are shown in Figs. 15 and 16 respectively. the graph shows the ups and downs in an individual's perfor-
mances embedded with specific factors of firing. The graphical
form can be analyzed to understand the standings of a person in
5.2.2. Result
case of firing.
After conducting the result, it is found that the performance
measure changes differently in case of variations in the factors
affecting the performance. The service length and the interval be- 5.3. Performance analysis
tween two consecutive firings have very less effect on the overall
performance of the firer. But the season has a remarkable impact on Performance analysis is important to take important decisions
the firing performance as it is related to sunlight and vision. The regarding the employment of a firer. The performance analysis
most influential factor among all these is the grouping and average module checks the available data related to a firer and can be

Fig. 15. Performance graph 1.

13
R.H. Chandan, N. Sharmin, M.B. Munir et al. Defence Technology xxx (xxxx) xxx

Fig. 16. Performance graph 2.

utilized for future prediction. Depending on the analysis, the 6. System development
decision-makers can take viable decisions and precaution
measures. 6.1. System architecture

5.3.1. Experimental setup The system has a total of four integral components.
For experimental purposes, we have taken data of eighteen
military personnel and put them in the prediction module of the 6.1.1. Component 1
system. For the analysis, we have taken the records of the Number The first components consists of a system that includes a good
of firings, Interval, Length of service, and Average Group from the quality camera and a wi-fi connection. The camera is connected to
individuals. The table (Table 9) below shows the corresponding the core system with wi-fi network to capture and fetch firing
predicted performances in relation to the provided data-set. images to the system.

6.1.2. Component 2
5.3.2. Result
The second components is the software itself which is inter-
The Results display the expected performance of the following
connected with other parts. It consists of the software module to
firing based on the firing history and other variables. Based on a
calculate and analyze the data collected from camera and the
firer's recorded results, the system anticipates the data by using
database and provide necessary results as its output.
machine learning methods. If the firer conducts firing more regu-
larly and records their continuous performances in the database,
the prediction's accuracy will only rise. However, if and only if the 6.1.3. Component 3
shooters can constantly fire, making very small in-size grouping The next one is the database system that provides a robust tool
rounds, the entire prediction process can produce superior results. to store necessary data.
In order to accurately determine the percentage of the performance
tables, the data set must also be robust. Good performers can be 6.1.4. Component 4
identified and hired as appropriate if this procedure is sorted ac- The fourth components is the output monitor display or the
cording to performance. printing machine (preferably the colored one) to show on screen
data or to print out the relevant results.

Table 9
Performance analysis.

Ser. Number of firings Inteval Season Length of service Average group Performance

1 2 33 0 11.00 7.35 69.5300


2 4 30 0 11.00 7.33 65.5700
3 1 3 0 11.00 7.34 66.2600
4 9 0 1.5 14.23 10.40 68.2673
5 8 0 1.5 12.22 8.16 69.0274
6 9 0 1.5 13.22 11.33 69.1585
7 8 0 1.5 12.81 18.29 53.5334
8 5 3 0 11.00 7.34 66.2600
9 4 3 0 11.00 6.34 66.6300
10 5 5 1.5 11.81 13.92 60.1855
11 4 5 1.5 12.22 17.66 56.2995
12 5 1 1.5 13.23 9.12 68.2450
13 3 1 1.5 12.81 25.56 58.1468
14 5 1 1.5 11.81 23.04 62.8377
15 4 1 1.5 12.23 17.81 61.3199
16 5 1 1.5 11.81 9.16 68.1895
17 8 0 1.5 12.23 16.95 62.9209
18 9 0 1.5 11.23 9.29 68.5262

14
R.H. Chandan, N. Sharmin, M.B. Munir et al. Defence Technology xxx (xxxx) xxx

Fig. 17. System architecture.

The functional system architecture is shown in Fig. 17 which project are mentioned here:
elaborates the ins and outs of the system precisely.
6.1.5. Computer vision 2
6.1.4.1. Platform used in the system. The platform of the whole The Computer Vision 2 or the CV2 library is used mainly for the
project is a programming language based application that can be image processing part. All the image processing modules required
used for processing the input and generate necessary outputs. The for image manipulation is taken from this library.
application is made using a robust language namely Python pro-
gramming language with its various modules and libraries. 6.1.6. Scikit image
The Scikit Image or skimage library is utilized in the structural
6.1.4.2. Front end. The Python tkinter library is used to build the similarity index measurement (SSIM) process.
front end of the system. The visuals are kept soothing to make the
users adaptable with the software. The built in modules of tkinter 6.1.7. Numerical Python
library is utlized fully the develop the total interface of the Numerical Python is used for various types of mathematical
software. calculations. This library is commonly known as numpy library.
Scikit Learn. Scikit Learn library provides an wide range of machine
6.1.4.3. Python libraries. Python's library is very extensive, offering learning methods. The Linear Support Vector Classifier (SVC)
a wide range of facilities. It contains various built-in modules that module is used in the error prediction part and the Linear Regres-
provide standardized solutions for many problems that occur in sion is used in the performance prediction part.
everyday programming. Some of these modules are explicitly
designed to address specific artificial intelligence and image pro- 6.1.8. Mathematical Plotting Library
cessing problems. The project contains a lot of Python modules, As the name says, the Mathematical Plotting Library is used to
however, the main libraries pertinent to the implementation of the generate various plots on the graphs.

Fig. 18. (a) Unit Selection Page; (b) Registration page.

15
R.H. Chandan, N. Sharmin, M.B. Munir et al. Defence Technology xxx (xxxx) xxx

Fig. 19. (a) Target compilation page; (b) Target with firing result.

Fig. 20. Individual firing results.

6.1.8.1. Back-end. For back-end data storage the SQLite database 6.2. UI screenshots
library of the Python programming language is used. It provides a
workable data storage system related to the project and provide a The following figures (Figs. 18e20) consists of screenshots of the
greater range of possibilities to upgrade it further. user interface for the titled project.

7. Conclusions
6.1.8.2. MATLAB FIS. To implement the FLS, the FIS module of
MATLAB software version 2021a has been used. The MAMDANI The paper presents three aspects related to small arms firing.
type interface is used to formulate the output of the system. The Firstly, it proposes an automated system to calculate points earned
system as an executable file is integrated with the Python module by the firer along with error detection using image processing and
afterwards to provide fuzzy outputs as a part of the project. machine learning techniques. Structural similarity index and
16
R.H. Chandan, N. Sharmin, M.B. Munir et al. Defence Technology xxx (xxxx) xxx

adaptive algorithm are used in the image processing parts to [7] Chandan MRH, Naim TA, Razzak MA, Sharmin N. Image processing based
scoring system for small arms firing in the military domain. In: Proceedings of
calculate firing scores. To find out the errors of the firer, machine
the 4th international conference on image processing and machine vision;
learning algorithms and ellipse recognition-based algorithms are 2022. p. 57e63.
used. Secondly, the system carries out performance analysis of the [8] Dzhilkibaeva N, Ahrens M, Laaksonen MS. Can performance in biathlon world
firer using fuzzy logic system. Thirdly, the system carries out per- cup be predicted by performance analysis of biathlon ibucup? Int J Perform
Anal Sport 2019;19(5):856e65.
formance prediction using linear regression model. [9] Ali F, Mansoor AB. Computer vision based automatic scoring of shoo-
The development of firing efficiency is a continuous process for tingtargets. In: 2008 IEEE international multitopic conference. IEEE; 2008.
military personnel. Units and training institutions also conduct p. 515e9.
[10] Ding P, Zhang X, Fan X, Cheng Q. Design of automatic target-scoring system of
practice firing regularly to upkeep the firing efficiency of their men. shooting game based on computer vision. In: 2009 IEEE international con-
The development of an artificial intelligence-based firing skill ference on automation and logistics. IEEE; 2009. p. 825e30.
evaluation system will play an important role in assessing the [11] Ye C, Mi H. The technology of image processing used in automatic target-
scoring system. In: 2011 fourth international joint conference on computa-
performance of a firer, analyzing his mistakes, and guiding him tional sciences and optimization. IEEE; 2011. p. 349e52. 730.
towards necessary correction. The performance predictions of a [12] Soetedjo A, Ashari MI, Mahmudi A, Nakhoda YI. Implementation of sensor on
firer will assist the officers to employ his men at the right place for the gun system using embedded camera for shooting training. In: 2014 2nd
international conference on technology, informatics, management, engi-
the right job. Moreover, the system will augment the trainer and neering & environment. IEEE; 2014. p. 69e74.
coach in performing their responsibilities. It will resolve the issue of [13] Lin Y-C, Miaou S-G, Lin Y-C, Chen S-L. An automatic scoring system for air
shortage of skilled-trainer, enable lifelong record keeping of firing pistol shooting competition based on image recognition of target sheets. In:
2015 IEEE international conference on consumer electronics. IEEE; 2015.
history, and decline the rate of fatigue and effort involved in a
p. 140e1.
manual or semi-automatic process. [14] Training Wing BMA. Field craft. 2000.
[15] Hatamleh KS, Khasawneh QA, Sawaqed L, Hassan MM, Yafawi S, Al-Shabi M.
Declaration of competing interest Evolutionary low-cost visual shooting practice system. In: 2015 10th inter-
national symposium on mechatronics and its applications (ISMA). IEEE; 2015.
p. 1e6.
The authors declare that they have no known competing [16] Issa AH, Hasan SD, Ali WH. Automation of real-time target scoring system
financial interests or personal relationships that could have based on image processing technique. Journal of Mechanical Engineering
Research and Developments 2021;44(2):316e23.
appeared to influence the work reported in this paper. [17] Strohm LS. An introduction to the sources of delivery error for direct-fire
ballistic projectiles. In: Tech. rep., ARMY RESEARCH LAB AB- ERDEEN
References PROVING GROUND MD WEAPONS AND MATERIALS RE- SEARCH; 2013.
[18] Zivkovic A, Hristov N, Jerkovíc DD, Bogdanovíc B, Milutinovíc J. Automatic
measurement of precision and accuracy from the hit pattern of small arms
[1] Borander AK, Voie ØA, Longva K, Danielsen TE, Grahnstedt S, Sandvik L,
using electronic target system. In: IOP conference series: materials science
Kongerud J, Sikkeland LIB. Military small arms fire in association with acute
and engineering. vol. 659. IOP Publishing; 2019, 012015.
decrements in lung function. Occup Environ Med 2017;74(9):639e44.
[19] Mendel JM. Fuzzy logic systems for engineering: a tutorial. Proc IEEE
[2] Ari
aujo D, Couceiro M, Seifert L, Sarmento H, Davids K. Artificial Intelligence in
1995;83(3):345e77.
sport performance analysis. Routledge; 2021.
[20] Ihalainen S, Kuitunen S, Mononen K, Linnamo V. Determinants of elite-level
[3] Lu G, et al. Prediction model and data simulation of sports performance based
air rifle shooting performance. Scand J Med Sci Sports 2016;26(3):266e74.
on the artificial intelligence algorithm. Comput Intell Neurosci 2022.
[21] Vapnik VN. The nature of statistical learning theory. new york: spring-verlag;
[4] Owusu G. Ai and computer-based methods in performance evaluation of
1995.
sporting feats: an overview. Artif Intell Rev 2007;27(1):57e70.
[22] Hong W-C, Pai P-F, Huang Y-Y, Yang S-L. Application of support vector ma-
[5] Li H, Zhang M. Artificial intelligence and neural network-based shooting ac-
chines in predicting employee turnover based on job performance. In: In-
curacy prediction analysis in basketball. Mobile Information Systems; 2021.
 Gieelczyk A, Chora ternational conference on natural computation. Springer; 2005. p. 668e74.
[6] wiklinski BC, s M. Who will score? a machine learning
[23] Hocking RR. Developments in linear regression methodology: 1959el982.
approach to supporting football team building and transfers. Entropy
Technometrics 1983;25(3):219e30.
2021;23(1):90.

17

You might also like