IMAGE CLASSIFICATION

Abstract:
The recognition and classification of the shapes such as Squares and Rectangles. Understanding the
identification of shapes in distinct images involves a deep process that has made usage of the recent
progress in neural networks which has brought the potential to train architectures to extract
features for this task. This project uses Convolutional Neural Network (CNN) techniques, a selection
of CNN architectures is evaluated to understand which is the best approach to extract features in
order to achieve outstanding results for the task. The results of the comparison emphasize the fact
that the accuracy and performance of the system improves, especially in the datasets which consist
of a large number of images. The proposed model achieves the highest training accuracy of 100%,
Validation loss 0.18% and Validation accuracy of 100%.

Approach:
The entire process is partitioned into different stages:
The preparation of training dataset,
Development of a CNN model,
Deep feature extraction for training the model,

Classification of the Shapes and

Single Image Prediction.

CNN MODEL
We propose a custom CNN-based model for recognizing rice leaf diseases. The model is designed
with a depth of several layers.
✓ input layer
✓ convolution layer 1 (Conv1)
✓ max pooling layer (Pooling1)
✓ convolution layer 2 (Conv2)
✓ max pooling layer 2 (Polling2)
✓ convolution layer 3 (Conv3)
✓ Three dense layers (Dense1, Dense2 and Dense3) and
✓ an output (softmax) layer

Hyper-Parameters of Our CNN-Based Model

➢ We use categorical cross entropy as the loss function while training our model.
➢ We run our model for a maximum of 15 epochs as there are no further improvements in
training and validation accuracies observed.
➢ Optimizer Adam is applied to optimize the loss function.
 ➢ We use 32, 64 and 32 (4×4) filters in Conv1, Conv2 and Conv3 layers, respectively, and
(2×2) max pooling in the pooling layers of our model.
➢ Batch Normalization is done individually at every hidden unit and Dropout rate is set to
10%.
➢ Activation functions such as ReLU experimented in each of the convolution layer of our
CNN model.

Performance Analysis
To evaluate the performance of our model, we consider the metrics such as Accuracy, Precision,
Recall and F1 score of each Shapes. Model achieves better results.

Conclusion
we have proposed a custom CNN-based model that can classify two shapes. Model is trained to
recognize the shapes such as Square and Rectangle. model achieves 100% accuracy on test
images and Validation loss is 0.18%. Moreover, Model is effective with respect to memory
storage due to its reduced number of network parameters.

PREPARED BY
Mohammed Musthafa CM

Date : 15/02/2022