This project presents a comprehensive approach for image classification using deep learning techniques, focusing on the integration of feature extraction from pre-trained convolutional neural networks (CNNs) and the application of custom CNN and MobileNet models. The dataset comprises images categorized into three classes, which are preprocessed and augmented to ensure robustness in model training. Initially, features are extracted using VGG16, VGG19, and InceptionV3 models, leveraging their pre-trained capabilities on the ImageNet dataset. These extracted features are then concatenated to form a comprehensive feature set, which is subsequently used to train custom CNN and MobileNet models. The models are evaluated using 10-fold cross-validation to ensure generalizability and robustness. Performance metrics, including accuracy, confusion matrix, and classification report, are utilized to compare the effectiveness of the models. The results demonstrate the superiority of combining feature extraction with advanced deep learning models, providing a highly accurate and reliable solution for image classification tasks. The project also includes detailed visualization of training and validation metrics to illustrate model performance. This approach highlights the potential of leveraging multiple pre-trained models for feature extraction, coupled with fine-tuned CNN architectures, to achieve state-of-the-art results in image classification.

Keywords: Convolutional Neural Networks (CNN), Feature Extraction, VGG16, VGG19, Data Augmentation, Machine Learning.