The Research on Remote Sensing Image Change Detection Based on Deep Learning

Abstract In the past, remote sensing image(RSI) change detection often ignored importance of shallow information with high resolution and fine grained features, there were problems missing rough boundary. This caused uncertainty edge pixels target loss small target. To solve such problem, we construct a new network under CDD data set for RSI detection. addition, in order to better integrate from bottom top layers, channel attention module is introduced conduct in-depth supervision. The testing results indicated that accuracy evaluation index reaches 88%, recall rate 86%, F1 value 83%, which solves pixel