Transfer learning-driven ShuffleNet with enhanced Res-UNet for computer-aided crop disease identification

The integration of deep learning into computer vision has greatly improved the efficiency and accuracy of crop disease detection. However, the lack of sufficient crop disease images limits recognition accuracy and the generalization of deep learning models. Enhancing the quantity and diversity of high-quality images is essential. This research proposes a transfer learning-driven Enhanced ShuffleNet model for computer-aided crop disease identification. Data acquisition was carried out by collecting images from crops such as grapes, potato, and tomato. The images were processed using anisotropic filtering to improve quality. An enhanced ResUNet model was employed for segmentation, improving generalization at multiple depths while retaining contextual details. Following segmentation, features such as color, enhanced median ternary pattern, and local phase quantization were extracted. The Enhanced Median Ternary Pattern captures texture information using local median values, enhancing edge and contrast detection. Finally, classification was performed using the transfer learning-driven EShuffleNet model, which incorporates an improved activation function to balance accuracy and computational efficiency. Experimental results show that the Enhanced ShuffleNet classifier achieved a sensitivity of 90% under 80% training and 92% under 90% training, demonstrating its effectiveness for accurate crop disease detection.