LodgeNet: Improved rice lodging recognition using semantic segmentation of UAV high-resolution remote sensing images

Abstract

• LodgeNet was used to obtain the information of different lodging degrees in rice. • LodgeNet achieved 97.30% accuracy in rice lodging semantic segmentation. • LodgeNet has a better effect than other networks on a small data set. Rice lodging not only causes difficulty in harvest operations, but also drastically reduces yield. Therefore, it is very important to identify rice lodging efficiently. For unmanned aerial vehicle (UAV) remote sensing images, this paper combines the advantages of dense block, DenseNet, attention mechanism, and jump connection on the basis of U-Net network to propose an end-to-end, pixel-to-pixel semantic segmentation method to identify rice lodging. And the method can process the input multi-band image. The accuracy of the model proposed in this paper was 97.30% on rice lodging images, which performed better than other comparison methods in the test. At the same time, it has good effect on small sample data set. The results show that it is feasible to use the improved U-Net network model to extract the lodging area of rice, which provide a useful reference for rice breeding and agricultural insurance claims.