Identification of Maize Leaf Diseases Based on MSDCNeXt Network

The complexity of the backgrounds, significant differences in the scale of the spots, and high similarity between the foregrounds and backgrounds result in low accuracy in identifying maize leaf diseases using existing methods. In the study, a multiscale deep neural network MSDCNeXt is proposed to identify maize diseases and solves this problems. In the MSDCNeXt network, each module constructs receptive fields of different sizes by stacking multiple layers of depthwise separable convolutional layers, improving the network's ability to express features of maize leaf diseases at different scales. CBAM attention mechanism is embedded in the MSDC block, enabling the network to more accurately locate key features of maize leaf diseases and effectively suppress interference from complex backgrounds. In addition, this study used data augmentation methods such as Mixup to further enhance the network's generalization ability.To verify the feasibility and effectiveness of the network in complex environments, we compared it with existing networks. The average identification accuracy of the network was 96.7%, and the proposed network outperformed the existing networks. This study demonstrates the capability for precise identification of images depicting diseases and pests affecting maize, thereby contributing to the development of diagnostic and management solutions for these afflictions in maize.