Landslide recognition in remote sensing images based on improved DDRNet
Landslide is a strongly destructive natural disaster,and the recognition and investigation of landslide disasters are an important basis for disaster prevention. The recognition accuracy and automation degree of traditional landslide recognition methods are low. Therefore,a landslide recognition algorithm based on a deep learning segmentation network was proposed in this paper. First,the dual resolution network model was used as the backbone network,and then the convolutional attention mechanism module was added to the backbone network to increase the model's ability to extract landslide features. Finally,the auxiliary loss function was added in the training stage to increase the model's ability to fit landslide features. The experiments show that the accuracy,recall rate,F1 score,and average intersection of union are all improved by about 5％,while the number of parameters is reduced by about two-thirds. It indicates that the proposed model has good landslide detection ability and can locate the landslide accurately and efficiently.
landslide recognitiondual resolution network segmentation modelconvolutional attention mechanism structureauxiliary loss function
万方数据
维普
