BiSeNet深度学习模型在四川阿坝州地区滑坡识别中的应用OA

Landslide has brought huge losses to human production and life.Its identification is the basis of landslide susceptibility,hazard and risk assessment,and provides data support for disaster management.Traditional visual interpretation requires expert experience,time-consuming and labor-intensive,while automatic classification method has low accuracy and cannot be applied on a large scale.Accompanying with the development of artificial intelligence,automatic landslide identification using high-resolution im-ages has become a hot research topic.In existing automatic recognition,the landslide is mainly composed of machine learning,with the application of the image as a data source less Google earth.In this paper,taking aba Tibetan and qiang autonomous prefecture in Sichuan province as the research area,using topog-raphy factor Google earth images as data sources,based on small samples,by using BiSeNet depth study method,landslide has been identified.First,the landslide sample set has been constructed based on Google Earth,and the landslide sample has been segmented and labeled by Labelme annotation tool.Sec-ondly,BiSeNet model has been used for semantic segmentation to train the landslide of sample set and la-bel set.Finally,the training model has been used to complete automatic landslide identification verification on the landslide data of the verification set.Combining with actual field verification,the accuracy of small sample identification method is 85.4%.It is showed that BiSeNet model based on image segmentation is effective for landslide recognition of small sample.