Based on the improved Deeplabv3 + remote sensing image semantic segmentation algorithm

As a new technology, remote sensing originated in the 1960s. In recent years, with the rapid progress of remote sensing satellite and computer hardware equipment, the number and ways of remote sensing image acquisition are increasing day by day. Due to the different sizes of remote sensing images, which are easily affected by the complex environment, this paper proposes a semantic segmentation algorithm based on the improved deeplabV3 +. In this paper, we use the lightweight network mobilenetv3 to extract features, then use the hollow pyramid structure to expand the receptive field, and then use the global context to enhance the fusion of features. Finally, bilinear interpolation method is used to upsample the output feature image to obtain the pixel level prediction segmentation map, so as to ensure the accuracy of image segmentation. Experiments show that the data provided by CCF big data competition (high-definition remote sensing image of a city in southern China in 2015) has good segmentation results. The total accuracy is achieved and the ratio of intersection and merging is achieved.