Semantic Segmentation of Open Pit Mining Area Based on Remote Sensing Shallow Features and Deep Learning

Abstract

Mineral resources are an important part of natural resources, and a good order of mining activities is an important prerequisite to ensure the safety of mining production, maintain the fairness of mining market, and steadily promote the construction of ecological civilization. Optical remote sensing image is one of the main carriers to reflect the mining activities of open-pit mining area. Deep learning technology is widely used in semantic segmentation of open-pit mining area. However, due to the complex surface environment of mining area, its segmentation accuracy needs to be further improved. In this paper, taking gaofen-2 optical remote sensing image as the data source, the remote sensing image sample set of open-pit mining area is constructed by manual annotation. Based on the sample set, the shallow texture features of the image are constructed, and part of the sample sets are put into the deep neural network for training. Combining the shallow texture features with the deep features of the deep neural network, a semantic segmentation model for pixel level open-pit mining area extraction is proposed by using U-net analysis model, and compared with the other two methods. The experimental results show that the overall accuracy of this method is 89.3%, and the average accuracy is 88.78%, which are better than the other two methods.