Research on Rural Feature Classification Based on Improved U-Net for UAV Remote Sensing Images

With the rapid development of economy and society, the planning and utilization of agricultural land is also crucial. Aiming at the problems of the current lack of Unmanned Aerial Vehicle Remote Sensing (UAV Remote Sensing) Image Dataset of rural surface features and the low efficiency and low automation of traditional surface feature classification methods, a rural surface feature detection segmentation method based on Improved U-net (U-net+) is proposed based on the semantic segmentation task in Deep Learning. This method integrates the Swin Transformer framework into the U-net infrastructure, which can not only effectively improve the operation efficiency of the model, but also easily extract feature information to improve the accuracy and generalization of the model. At the same time, the UAV Remote Sensing image is used as the data source to produce a general UAV Remote Sensing Image Dataset of rural features: Air Dataset. In order to verify the advantages of U-net+, this study uses two UAV Datasets, the self-made Air Dataset and the famous Semantic Drone Dataset, to operate in U-net+. The experimental results show that the U-net+ proposed in this paper achieves 60.3% of mean Intersection-over-Union (mIoU) in the Air Dataset, which is better than U-net in PA, detection efficiency and model generalization ability; The Semantic Drone Dataset mIoU achieves 39.7%, which is also better than U-net According to the comparative analysis of the two Datasets, the segmentation accuracy of the two models for house buildings and grassland vegetation is relatively high, and the U-net+ is better. Therefore, in actual production and life, the U-net+ is more suitable for the rural terrain scene segmentation task mainly based on “Fine Operation”.