Deep Learning Image Classification with CIFAR-10 Dataset Using CNN (with Transfer learning using MobileNetV2 architecture, pre-trained on the ImageNet dataset, for object detection on CIFAR-10 )
Introduction: This project aims to implement an image classification model using deep learning frameworks for the CIFAR-10 dataset. The dataset consists of 60,000 color images with dimensions of 32x32, divided into 50,000 training images and 10,000 test images. The images are categorized into ten classes. The project follows a convolutional neural network (CNN) architecture with specific layers and activation functions.
Dataset: The CIFAR-10 dataset can be obtained from its official webpage and used for training and testing the models. Optionally, only the first category (1_data_batch) can be used as training data, if resource constraints apply.
Model Architecture: The CNN model comprises the following layers:
1- Convolutional layer: Kernel size 3x3, Output channels 7, Activation function: ReLU. 2- Convolutional layer: Kernel size 3x3, Output channels 9, Activation function: ReLU. 3- Max Pooling layer: Kernel size 2x2. 4- Dropout layer with 30% probability. 5- Fully connected layer with the number of neurons equal to the number of classes.
Training: The model is trained using the Adam optimizer, categorical cross-entropy loss, and a learning rate of 0.01 (batch size 32). The training process includes validation data to monitor convergence. The model's accuracy is evaluated using the test dataset.
Experimentation: Several modifications are applied to the base architecture and parameters to analyze their impact on performance:
1- Varying the number of convolutional layers from two to four. 2- Changing the kernel size of the first convolutional layer to 5x5 and 7x7. 3- Replacing the first and second convolutional layers with a single 5x5 kernel layer. 4- Modifying the learning rate of the optimizer to 0.0004, 0.0003, and 0.0001. 5- Trying alternative activation functions like leaky ReLU and tanh. 6- Swapping the optimizer to SGD and Adam. 7- Adding Batch Normalization as the first layer of the network. 8- Changing the batch size to 4 and 128.
Visualizing Convolutional Layers: After training, visualize the output of the first convolutional layer for a sample test image. Display each channel as a grayscale image. Additionally, visualize the kernel weights of one of the channels.
Transfer Learning with MobileNetV2: Transfer learning is implemented with MobileNetV2 architecture, pre-trained on the ImageNet dataset, for object detection on CIFAR-10. Linear layers are added with 10 output neurons for classifying the CIFAR-10 categories after the feature extraction layers. The convergence speed and accuracy are compared with the base model.