High Resolution Remote Sensing Image Classification Based on Deep Learning U-Net Model

Abstract

The higher the resolution, the clearer the content of the image will be. With the continuous improvement of remote sensing image(RSI) resolution in recent years, the detailed information contained in the image becomes clearer and richer, which is an important breakthrough for remote sensing research. Deep learning U-Net model(DLUM) can build training model from shallow level to deep level in data learning, increase model training parameters, and guide classification decision more efficiently and accurately. This paper studies the classification method of high-resolution RSI based on DLUM, which can reduce the network computing time, improve the classification accuracy, and then improve the classification results of high-resolution RSI. This paper designs the DLUM, understands the high-resolution RSI solution of u-net model classification, and preprocesss the experimental high-resolution remote sensing data. Firstly, the experiment carries on coordinate system transformation and visualization processing, then normalizes the data, then enhances the data by some image processing methods, and finally transforms the data into mask to express the characteristics of each kind of figure. This paper studies the advantages of the high-resolution RSI classification method based on DLUM through experiments, and compares it with the traditional RSI classification method intuitively through chart analysis method to analyze its accuracy. The experimental results show that the classification method of high-resolution RSI which is added by DLUM has obvious superiority. On the ground samples, the accuracy of high-resolution RSI classification of basic DLUM is 89.10%, while the accuracy of traditional classification method is 79.80%.