Fast and Efficient CNN Architectures for Automated Soybean Leaf Disease Classification

Soybean leaf diseases significantly affect crop yields in Indian agriculture. Timely detection is crucial for managing these diseases effectively-‎ly. While traditional CNN models deliver high accuracy, they have significant computational demands and slower speeds, limiting real-time ‎use. This study aims to create faster CNN models that effectively identify eight common soybean leaf diseases while being easier to train. ‎The method uses a Separable-CNN with a Global Average Pooling (GAP) layer to reduce parameters and computing needs. The dataset ‎used to train the Fast CNN and Enhanced CNN models included 40,000 images across eight disease categories. These models underwent ‎evaluation using validation and test datasets with varying splits: 70-20-10, 65-25-10, and 60-30-10. The Enhanced CNN achieved 98.14% ‎test accuracy, outperforming models like VGG16, ResNet152V2, and InceptionV3. The Fast CNN achieved 95.70% accuracy with reduced ‎training time, making it suitable for real-time deployment. The confusion matrix showed high accuracy with few errors. The new CNN ‎models accurately identify soybean leaf diseases and work efficiently. These models could help manage diseases quickly and improve AI ‎tools in agriculture‎.