Leveraging Spatial and Channel Attention in VGG16 for Efficient Tomato Leaf Disease Detection

Plant diseases result in devastating agricultural losses despite the vital role that agriculture plays in the global economy. To combat such loss, Worlds economic powers have invested trillions of dollars in plant disease management. Especially, Tomato leaf diseases such as ’Bacterial spot’, ’Target Spot’ etc. Currently, various Convolutional Neural Network (CNN) based architectures have proven effective in detecting small-scale plant leaf diseases. Our proposed architecture, referred to as Visual Geometry Graph with 16 layer depth or VGG16 SpatialChannelAttention (VGG16_SCA), is a cutting-edge (CNN) design that is especially suited for image classification applications. The VGG16_SCA model combines a pre-trained VGG16 model with additional Spatial and Channel Attention modules, enhancing its capabilities for image classification tasks. The VGG16 backbone extracts fundamental image attributes such as edges, corners, colors, and shapes. Meanwhile, the carefully designed Spatial and Channel Attention modules capture relationships within and between channels, allowing the model to emphasize crucial features and disregard irrelevant ones. Finally, The VGG16_SCA model is computationally efficient while maintaining a comparatively simpler structure compared to other deep CNN commonly used for image classification tasks. Due to its suitability for applications with limited computational resources. Our architecture proposes an accuracy of 99.04% while identifying tomato leaf diseases. To understand model performance, our research study provides comparison studies, expert analyses, and evaluation measures. Based on the results, it can be said that the VGG16 SpatialChannelAttention (VGG16_SCA) architecture, when combined with image processing and statistical analysis, is capable of accurately diagnosing different degrees of tomato plant diseases.