CSASNet—A Crop Leaf Disease Identification Method Based on Improved ShuffleNetV2

In identifying crop leaf diseases, Due to the complex nature of the disease symptoms. There may be variations in disease symptoms with similar characteristics and similarities in disease symptoms with different elements. This can make it challenging to differentiate between various diseases. CSASNet is a hybrid classification model proposed in this paper that combines the attention and multiscale feature fusion mechanisms. The model first incorporates the multiscale feature fusion module atrous spatial pyramid pooling (ASPP) into the ShuffleNetV2 network structure. This enriches the network with disease-specific multiscale feature information. Additionally, the model combines the special group-wise enhance (SGE) attention mechanism module to enhance the weight of disease spot feature information. Lastly, the leaky ReLU function replaces the original ReLU activation function. This allows the model to reduce damaging feature loss during training. The paper presents a design of multiple cross-validation experiments for comparison. The experimental results suggest that the improved model was used for disease leaf identification and showed an accuracy improvement on different crops. Compared to Convnext and MobileNetV2, the CSASNet model demonstrates higher recognition accuracy.