Research on Maize Recognition at Field Parcel Scale Based on Improved U-Net Network

Accurate crop identification and classification is the core issue of crop monitoring, which is substantial importance to master grain production and realize scientific management. To enhance the precision of maize recognition at field parcel scale, this study took maize as research object, introduced Convolutional Block Attention Module based on U-Net, and proposed the maize recognition method at field parcel scale based on improved U-Net network. The comparative analysis between the improved network's recognition outcomes and those of the original U-Net revealed that the upgraded version demonstrated the swiftest training pace and the most rapid descent in the loss function, achieving stability sooner. The improved U-Net network's recognition accuracy was 88%, which represented a 5% increase over U-Net. The improved U-Net network recognized crops more accurately, which offers innovative insights for improving the accuracy of crop recognition at field parcel scale.