Remote Sensing Image Recognition and Classification Based on Complex Networks

This paper presents a complex network-based feature extraction method designed to address the diversity and complexity of remote sensing images to enhance classification accuracy. The proposed method comprises three key components: 1) texture analysis of remote sensing images in various settings, showing differences in the distribution of edge weights within the first regular network according to texture categories, where rougher textures require higher thresholds to form the network; 2) feature extraction through a combination of static statistics and threshold evolution within the complex network, leading to the creation of a robust network model; 3) a convolutional neural network is utilized to classify and predict the processed remote sensing images, and the results are compared to the original image categorization. The experimental results demonstrate that images processed via complex networks exhibit higher classification accuracy. This method performs superior classification and provides new insights into remote sensing image classification and recognition.