Performance evaluation of an enhanced shufflenet CNN for multi-crop leaf disease classification using fine-tuned parameters

Plant leaf diseases can reduce crop quality and cause big losses to farmers. Many current models used to detect these diseases do not work well when images have poor lighting or messy backgrounds. This study enhances the ShuffleNet CNN model to detect leaf diseases in different crops like capsicum, rice, corn, tomato, and citrus. Leaf images were taken using a Kinect camera, which gives clearer images in farm conditions. The improved ShuffleNet model was trained with fine-tuned settings: 0.010 learning rate, 64 batch size, 50 training rounds, and the Adam optimizer. It achieved a high accuracy of 91.94%, performing better than other models like ResNet50 and DenseNet201. The model also showed strong results in precision, recall, and F1 score. In conclusion, the enhanced ShuffleNet is a reliable and fast tool for detecting leaf diseases in many crops and is useful for smart farming.