Identification Method of Cotton Leaf Diseases Based on Bilinear Coordinate Attention Enhancement Module

Abstract

Cotton is an important cash crop. Cotton diseases have a considerable adverse influence on cotton yield and quality. Timely and accurate identification of cotton disease types is important. The accuracy of cotton leaf disease identification is limited by unpredictable factors in natural settings, such as the presence of a complex background. Therefore, this paper proposes a cotton leaf disease identification model based on a bilinear coordinate attention enhancement module. It reduces the loss of feature information by bilinear coordinate attention embedding feature maps spatial coordinate information and feature fusion. Hence the model is more focused on the leaf disease region and reduces the attention to redundant information such as healthy regions. It also achieves the precise localization and amplification of attention to the leaf disease region through data enhancement, which effectively improves the recognition accuracy of cotton leaf diseases in a natural setting. By experiments, the identification accuracy of the proposed model is 96.61% and the parameter size is 21.55 × 106. Compared with other existing models, the identification accuracy of the proposed model is greatly improved without increasing the parameter size. This study can not only provide decision support for the timely diagnosis and prevention of cotton leaf diseases but also validate a paradigm for the identification of other crop leaf diseases.

Keywords

• natural environment
• identification of cotton leaf diseases
• bilinear coordinate attention mechanism
• data enhancement
• ResNet