EfficientNet–Fuzzy ShCNN for Multi-Level Cotton Leaf Disease Classification under Complex Environments

Cotton diseases reduce crop yield and fiber quality, leading to significant agricultural losses. Early detection plays a key role in effective disease control and crop management. The presence of visible infection patterns on cotton leaves makes it possible to identify these diseases at an early stage. This study proposes an effective model for cotton leaf disease classification using a hybrid EfficientNet–Shepard Convolutional Neural Network (EffNet_ShCNN). The input images are first processed using a Kalman filter for noise reduction and enhancement. Subsequently, leaf segmentation is performed using Fuzzy Local Information C-Means (FLICM) clustering. Multiple features, including shape, color, texture, CNN features, Local Gabor Binary Pattern (LGBP), Spider Local Image Feature (SLIF), and Weber Local Descriptor (WLD), are then extracted. These features are used for disease detection to classify leaves into healthy and unhealthy categories. The proposed model integrates EfficientNet and ShCNN with fuzzy-enhanced layers to improve classification robustness. Finally, the model performs multi-class classification into three disease categories: aphids, bacterial blight, and target spot. Experimental results demonstrate that the proposed method achieves a maximum accuracy of 91.70%, sensitivity of 90.40%, and specificity of 92.90%, indicating its effectiveness under complex environmental conditions.