A Method for Sugarcane Disease Identification Based on Improved ShuffleNetV2 Model

Abstract Rapid and accurate identification of sugarcane diseases is an important way to improve sugarcane yield. Therefore, this study proposes an improved model based on ShuffleNetV2 network (Im-ShuffleNetV2) for sugarcane disease identification. Firstly, we incorporated the ECA (Enhanced Channel Attention) attention mechanism into ShuffleNetV2, enhancing the network's ability to extract features and detect sugarcane lesion areas. Secondly, a new multi-scale feature extraction branch and Transformer module have been introduced, further improving the independent learning ability of the network. Finally, a large number of numerical results have demonstrated the advantages of the proposed model in terms of parameter size and sugarcane disease identification accuracy. Just as Im-ShuffleNetV2 only has a parameter of 0.4MB, it has significant advantages over parameters such as EfficientV2-S (55.6MB), MobileNetV2 (8.73MB), MobileViT XX small (3.76MB), FasterNetT2 (52.4MB), AlexNet (55.6MB), and MobileNetV3 Large (16.2MB). In addition, compared with the ShuffleNetV2 network, the accuracy has improved by 3.4%. This model not only improves the accuracy of sugarcane leaf disease detection, but also demonstrates the advantage of lightweight, providing valuable reference for future research in the field of sugarcane.