A Compact CNN Architecture for Detection and Classification of Cotton Leaf Diseases

This study presents a comparative analysis of convolutional neural network (CNN) architectures for the classification of cotton leaf diseases, a critical effort in agricultural disease prevention and control. The challenge of deploying complex models on devices with limited computational resources is addressed by proposing a lightweight CNN architecture that ensures a balance between model complexity and detection accuracy. In order to reduce computational costs while maintaining the essential features necessary for accurate disease detection, network pruning, depth wise separable convolutions, and point wise convolutions are employed. Furthermore, the model is capable of performing well with minimal input data by utilizing transfer learning with pre-trained weights to improve learning efficiency. The proposed architecture is assessed in comparison to renowned models, such as ResNet50, ResNet18, and VGG16. ResNet50 exhibited the highest performance among these models, with an accuracy of 97.8%. Consequently, it is the most appropriate model for cloud-based services that require real-time monitoring and disease detection. The lightweight CNN model, despite being less accurate than ResNet50, provides advantages in low-power environments, obtaining competitive result of 97.5% accuracy. This study emphasizes the potential of CNNs to automate the diagnosis of agricultural diseases, providing scalable solutions for resource-limited environments, and establishing the groundwork for future developments in real-time monitoring systems for agriculture.