Contour Based Target Recognition in Aerial Images

Target recognition for aerial images is an important research issue in remote sensing applications. Many feature-based recognition methods have been introduced for target recognition. Nevertheless, these methods have their limitations when considering the large amount of data provided by satellite imagery. In this paper, we explore several techniques for target recognition in aerial images with a contour matching approach. Contours in our approach are detected by a contour grouping strategy and described by edge potential function, which provides an attraction field for edges with similar curves. In this sense, target recognition can be formulated as an optimization problem. An improved chemical reaction optimization (CRO) algorithm is proposed in this paper to deal with the target matching problem.

Experimental results demonstrate the robustness and high efficiency of our approach over the state-of-the-art evolutionary algorithms, which include the original CRO, predator-prey biogeography-based optimization, an improved version of brain storm optimization, artificial bee colony, quantum-behaved particle swarm optimization, a self-adaptive differential evolution algorithm, and stud genetic algorithm. In addition, several case studies regarding remote sensing are also presented. The results show that the proposed method is capable of improving the application ability of recognizing target in aerial images