关键词: 生产建设项目, 变化检测, U-Net++, 人机协同

Abstract: Disturbed patch interpretation of production and construction projects in soil and water conservation is mostly realized by artificial visual interpretation of remote sensing images. In the actual work process, there are some problems such as low efficiency, high cost and strong subjectivity.This paper proposes a human-machine collaborative intelligent extraction framework for production and construction project disturbance patches, which combines intelligent extraction model with remote sensing supervision cooperation platform. The change detection dataset is constructed by elements annotation, data enhancement and other means. And then, the improved U-Net++ model is used to carry out the intelligent extraction of production and construction project disturbance patches. The results show that the average accuracy of the model is 79.59%, and the area recall rate of the model is 80.90%.In addition, aiming at the problems that the model easily extracts the pseudo-variation or cloud obscured region incorrectly, patch fragmentation, or irregular contour boundary, a distributed parallel collaborative interpretation platform is built on the basis of automatic extraction results. The platform can realize the functions of adding, deleting, creating, quality inspection, and so on.The final results are fed back to the model as new samples to further improve the performance of the model. Thus form a virtuous cycle between the sample and the model, and improve the actual work efficiency.

Key words: production and construction projects, change detection, U-Net++, human-machine collaboration