Classification Method of Remote Sensing Image Objects Based on Deep Learning Semantic Segmentation Network

Abstract

The invention discloses a remote sensing image object classification method based on a deep learning semantic segmentation network. Firstly, various objects in the remote sensing image are marked pixel by pixel, and a remote sensing object annotation image library is constructed as a training label. Subsequently, the present invention designs a method for constructing a multi-scale feature map group based on texture and structural features, and combines the feature map group and the original image as the input of the deep learning network. In addition, the present invention designs a method based on the deeplab algorithm. The improved network structure of the fully convolutional network, parameter training is carried out through convolution and deconvolution, and finally the wide-range remote sensing images are overlapped and segmented, and after classification, they are merged to obtain the final classification results of wide-range remote sensing images. It can efficiently and quickly realize the pixel-level classification of various ground features in high-resolution remote sensing images, simplify the complex process of traditional classification methods, and achieve good segmentation and classification results.

Description

technical field [0001] The invention belongs to the technical field of intelligent classification of remote sensing images, and more specifically relates to a classification method of remote sensing ground objects based on a fully convolutional semantic segmentation network under the requirement of ground object interpretation. Background technique [0002] Remote sensing image classification is currently widely used in various military and civilian applications such as land surveying, satellite law enforcement, and regional investigation, and has achieved good application results and has great market development potential. With the gradual increase of satellite load and data volume, in the research of high-precision remote sensing classification, especially when facing large-scale (national or global scale) surface classification, the traditional manual calibration method is difficult to support the explosive growth of tasks and demands. Therefore, it is an important task w...

AI Technical Summary

Problems solved by technology

With the gradual increase of satellite load and data volume, in the research of high-precision remote sensing classification, especially when facing large-scale (national or global scale) surface classification, the traditional manual calibration method is difficult to support the explosive growth of tasks and demands. Therefore, it is an important task with far-reaching significance to study how to use artificial intelligence to realize the intelligent automatic processing of remote sensing images.

Images