Multi-Scale Segmentation Method of Remote Sensing Big Data Image Using Deep Learning

Remote sensing image (RSI) segmentation is an effective method to interpret remote sensing information and an important means of remote sensing data information processing. Traditional RSI segmentation methods have some problems such as poor segmentation accuracy and low similarity difference measurement. Therefore, we propose a multi-scale segmentation (MSS) method for remote sensing big data image. First, the segmentation scale of RSI is divided, and the quantitative value of histogram band is used to calculate the similarity index between different objects; Second, the parameters in the same spot are improved based on the maximum area method to determine the shape factor of RSI; Finally, the object closure model is established to clarify the region conversion cost, and the RSI is dynamically segmented based on Multi-scale convolutional neural networks; The MSS algorithm of RSI is designed, and the MSS method of RSI is obtained. The results show that the maximum similarity difference measure of the proposed method is 0.648, and the similarity difference measure always remains the largest. The maximum recall of RSI is 0.954, and the highest recall is 0.988, indicating that the RSI segmentation accuracy of the proposed method is good.