SCECA U-Net crop classification for UAV remote sensing image

Timely and accurate extraction of crop planting units plays a crucial role in crop yield estimation, soil management and disaster warning. For small-scale crop cultivation systems, accurate mapping of crop types is challenging due to the presence of heterogeneous and mixed pixels, while the field area is small and crop types are very diverse. Considering that traditional remote sensing methods are difficult to accurately extract crop features due to factors such as cost and sensitivity to weather conditions. Therefore, we select unmanned aerial vehicle remote sensing images of Majiaoba Town, Mianyang City, Sichuan Province, China, aiming at identify and classify various crops in the region, including soybeans, corn, rice, ginger and walnuts. We propose using the Visual Geography Group Network-16 architecture as the main framework, enhancing it by integrating efficient channel attention modules and spatial and channel reconstruction convolution modules into the U-Net model. It is worth noting that the improved model achieved an impressive F1-score of 0.78, emphasizing its effectiveness in crop classification. Experimental results show that the method achieves a comprehensive classification accuracy of 79.44% and an average cross-combination rate of 0.64. Our approach outperforms traditional U-Net, Pyramid Scene Parsing Network, and deeplabv3+ models in terms of classification accuracy and metric, highlighting its advantages in crop identification.