Enhanced Plant Disease Detection Using a Custom CNN with Advanced Feature Extraction Techniques

Prompt and precise identification of plant diseases is essential to guarantee agricultural output and food security. Conventional techniques that depend on visual examinations are inefficient, based on personal judgment, and frequently cannot be expanded to larger scales, emphasizing the necessity for automated alternatives. Recent advancements in convolutional neural networks (CNNs) have shown promise in addressing these challenges, offering high accuracy in image classification tasks. However, existing CNN architectures often require extensive computational resources, limiting their real-time application and scalability. Ensemble methods, while improving accuracy, also increase model complexity and deployment difficulty. To address these issues, we propose a custom CNN architecture specifically designed for plant disease detection. Our model includes advanced features like depthwise separable convolutions, dilated convolutions, and an attention mechanism to improve the efficiency of feature extraction and focus on essential parts of the image. These enhancements ensure high classification performance without imposing excessive computational demands. Our experimental results demonstrate the superiority of the custom CNN, achieving a remarkable validation accuracy of $\mathbf{9 9 . 4 4 \%}$, precision of 0.98, recall of 0.97, and an $F 1$ score of 0.97. By the fifth epoch, the model’s training process shows a consistent decrease in loss, with the training loss dropping to 0.0245 and the validation loss to 0.0196. These results underscore the model’s ability to generalize effectively to unseen data while maintaining computational efficiency suitable for real-time applications. The proposed custom CNN significantly outperforms traditional CNN architectures and ensemble methods in plant disease detection.