Remote sensing image semantic segmentation combining UNET and FPN

The traditional remote sensing image segmentation method is inefficient and the segmentation fineness is not enough in complex scenes. The UNET model is well-known for its good segmentation effect, but it does not perform well for the smaller objects contained in the image and the edge segmentation of larger objects. In order to solve this problem, a method combining UNET structure with FPN structure is proposed in this paper to improve the ability of UNET model to integrate multi-scale information. At the same time, the BLR loss function which can better capture the edge of the target edge is used to improve the segmentation effect of UNET model on the target boundary. The experimental results show that the method used in this paper effectively improves the accuracy of semantic segmentation and alleviates the problem of poor edge segmentation of small-scale targets and large-scale targets. The target edge segmentation can be more accurate to achieve better segmentation results.