Technical Seminar (Wildfire)

You might also like

Download as pptx, pdf, or txt
Download as pptx, pdf, or txt
You are on page 1of 16

Technical Seminar Presentation on

A Deep Learning - Based Experiment on


Forest Wildfire Detection in Machine Vision
Course
Department of Artificial Intelligence and Machine
Learning
Contents

• Introduction

• Wildfire Image Classification Module

• SVM Classification

• Reduce-VGG Net Model

• Wildfire Region Detection Module

• Vibe Algorithm

• Experimental Results

• Future Directions

• Conclusion

• References
INTRODUCTION
• A deep learning-based experiment on forest wildfire detection in a machine vision course. The experiment
involves feature extraction from wildfire images using digital image processing techniques.
• The extracted features are normalized to accelerate the convergence speed of the
algorithm. The normalized features are input into SVM and Reduce-VGG Net for classification, and
the performance is compared.
• The SVM classifier is a popular machine learning algorithm that can realize nonlinear segmentation
of feature vectors. The Reduce-VGG Net model is proposed to reduce the training parameters and
training time in the VGG Net model.
• Wildfire region detection module is a two-stage deep neural network model for detecting wildfire
regions. The first stage detects moving objects using the Vibe algorithm. The second stage classifies
each block of the moving object as a wildfire block or non-wildfire block. The paper compares the
proposed method with well-known wildfire detection algorithms and state-of-the-art object detection
algorithms.The proposed method performs better than other methods in detecting wildfire regions.
• Preprocessing: The module starts by extracting video image frames using the OpenCV module, which offers
robust video editing capabilities for image processing tasks.
• Feature Extraction: Following preprocessing, the module focuses on extracting essential features from the
images, including colour, texture, and shape features. This step is crucial for subsequent classification tasks.
• Algorithm Selection: The module proposes a novel wildfire image classification algorithm based on traditional
machine learning methods like SVM (Support Vector Machine) and a specialized algorithm based on Reduce-
VGG Net.
• Model Training: Students are required to compare the performance of the proposed Reduce-VGG Net
algorithm with the traditional SVM approach. This involves training the models on the extracted features and
evaluating their accuracy.
• Evaluation: The accuracy of the classification models is used as a metric to assess their performance. The
module aims to enhance students' understanding of image preprocessing, feature extraction, and classification
techniques in the context of wildfire detection.

Wildfire Image Classification


Module
Fig :- The flow chart of the SVM
classification

SVM
CLASSIFICATION
The normalized features are input into an SVM classifier, which can realize nonlinear segmentation of feature vectors.
The performance of the SVM classifier depends on the selection of the kernel function, which simplifies the number
of inner product calculations, reduces running time, and converts the inner product of high-dimensional space to a
low-dimensional space. The paper uses the Lib svm package to establish the classification model and compares the
performance of different kernel functions, including Radial Basis Function (RBF), Polynomial Kernel, and Sigmoid
kernel. The optimal parameters are selected based on cross-validation.
• The Reduce-VGG Net model proposed in the paper is a
modified version of the VGG-16 network, which consists of 13
convolutional layers and 3 fully connected layers.
• The model is designed to reduce the training parameters and
training time in the VGG Net model.
• The authors transfer the weight parameters obtained from the
training set of the network to the wildfire image set and fine-
tune the VGG16 for classification.
• The model is trained using stochastic gradient descent (SGD)
and Momentum, with the initial learning rate set to 0.001,
which is updated using the exponential decay method.
• The Reduce-VGGNet model performs better than the SVM
algorithm in wildfire image classification.

Reduce-VGG Net Model


Fig :- The structure of Reduce-VGG Net.
Wildfire Region Detection
Module
• Wildfire region detection involves two stages: moving object detection using the Vibe algorithm and
classifying each block as wildfire or non-wildfire.
• Deep CNN architectures are utilized for classification due to their capacity to extract highly
representative features.
• New CNNs are designed to extract both spatial and temporal features to address the varying scales
of forest wildfires.
• The Vibe algorithm is employed for background modeling and foreground detection in wildfire
region detection.
• The proposed method, integrating optimized CNNs and the Vibe algorithm, outperforms traditional
machine learning algorithms like SVM in wildfire image classification accuracy.
Fig :-The framework of forest
wildfire region detection.
• Vibe algorithm initializes the background model with the first
frame of the video.
• It randomly selects neighbouring pixel values as sample values
for each pixel.
• The algorithm determines if a pixel is part of the background
by comparing distances to values in the sample set.
• Background modeling involves storing a sample set for each
background point.
• The algorithm calculates distances between new pixels and
values in the sample set to classify them.

Vibe
Algorithm
EXPERIMENTAL
RESULTS
Fig :- The structure of temporal CNN network

Fig :- The results of spatial feature detection Fig :-The results of temporal feature detection
based on the optimized CNN. based on the optimized CNN.
RESUL
TS

Fig :-The experimental results of different methods.


FUTURE
DIRECTIONS
• Improve Wildfire Image Classification

• Optimize Wildfire Region Detection

• Integrate Wildfire Detection with Real-time Systems

• Improve Wildfire Detection in Noisy Conditions

• Collaborate with Wildfire Authorities

• Continuous Research and Development


CONCLUSIO
N
• experiment on forest wildfire detection using deep learning in
a machine vision course. It involves two main modules:
Wildfire Image Classification and Wildfire Region Detection.
• The experiment evaluates accuracy for image classification
and uses precision, recall, and accuracy for wildfire region
detection.
• The dataset includes wildfire and non-wildfire images, with
manual annotation for wildfire regions. Results show the
effectiveness of normalization and the superiority of the
Radial Basis Kernel function in SVM classification. The
Reduce-VGG Net network achieves 91.20% accuracy on the
test set.
REFERENCE
S
[1] A. Kumar, A. Kaur, and M. Kumar, ‘‘Face detection techniques: A review,’’ Artif. Intell. Rev., vol.
52, no. 2, pp. 927–948, 2019.

[2] M. J. Sousa, A. Moutinho, and M. Almeida, ‘‘Wildfire detection using transfer learning on
augmented datasets,’’ Exp. Syst. Appl., vol. 142, Mar. 2020, Art. no. 112975.

[3] L. Wang, Y. Zhang, and X. Xu, ‘‘A novel group detection method for finding related Chinese
herbs,’’ J. Inf. Sci. Eng., vol. 31, no. 4, pp. 1387–1411, 2015.

[4] B. Tang, J. Kong, and S. Wu, ‘‘Review of surface defect detection based on machine vision,’’ J.
Image Graph., vol. 22, no. 12, pp. 1640–1663, 2017.
Thank you

You might also like