A Dynamic Effective Class Balanced Approach for Remote Sensing Imagery Semantic Segmentation of Imbalanced Data

The wide application and rapid development of satellite remote sensing technology have put higher requirements on image segmentation methods. Because its characteristics large size, data volume, complex background, not only are the traditional methods difficult to apply effectively, but based deep learning faced with problem extremely unbalanced between categories. In order solve this problem, first all, according existing effective sample theory, calculation method in context semantic is firstly proposed highly dataset. Then, a dynamic weighting concept proposed, which can be applied images. Finally, applicability different loss functions network structures verified self-built Landsat8-OLI image-based tri-classified forest fire burning area dataset LoveDA dataset, for land-cover segmentation. It has been concluded that algorithm enhance minimal-class accuracy while ensuring overall performance multi-class tasks two tasks, including land use cover (LULC) addition, significantly improves recall by as much about 30%, great reference value research