A novel feature extraction approach for remote sensing image based on the shape-adaptive neighborhood

Abstract

Feature extraction is a significant procedure for target recognition and classification of remotely sensed images. In this paper, the previous feature extraction methods were considered to consist of three layers: the abstract layer, the methodology layer and the feature layer. A new feature extraction approach based on the shape adaptive neighborhood (SAN) in the abstract layer was proposed. Firstly, the heterogeneity based on the color characteristics was defined to determine the SAN for each pixel. Then all the color, texture and shape features were extracted from each SAN, and fused together by a feature level fusion method. In the experiment, the features were used to execute the classification work. As the results showed, most of the SANs contained enough pixels for the texture and shape analysis, and the total precision was 0.9354.