Multi-scale analysis for remote sensing target recognition

Abstract

To solve the problem that the features of a target change with the image resolution in target recognition, a multi-scale analysis of target recognition was provided in this paper. First, the multi-scale features of a target were analyzed based on mixed pixels in multi-resolution images. At different scale, the features of a target had different behaviors that were related to the proportion of mixed pixels in the target. Second, the structural differences of the target were analyzed and illustrated from the experiments of multi-scale target recognition. Last, the concept of "dominating class" was proposed based on the multi-scale analysis. The experimental results show that the shape and structural features change with scale of the image and large scale leads big change. This paper studies the target features from multi-scale images and provides guidance on image target recognition.