Aircraft Detection Research In Remote Sensing Images Based On Visual Saliency

With the development of remote sensing technology, the quality of remote sensing images has improved significantly, which makes researches on object detection in these images more and more popular. For all those ground targets, research on aircraft detection is of great importance. In the military field, aircraft is one of the most important tools of modern warfare and plays a vital role in the progress of a battle. So it is quite necessary to find out the deployment and real-time dynamic of enemy aircrafts, and we may strike them precisely to win the war. In the civil filed, as an indispensable modern traffic tool, aircraft has greatly enhanced the quality of people's life. We can archive automation management of airport by aircraft detection in remote sensing images, which can both save manpower and improve the safety.Nowadays plenty of researches have been carried out on aircraft detection in remote sensing images at home and abroad, with profitable results been archived. But because of the huge amount of information and complex backgrounds in the remote sensing image itself, along with the disturb of image noise and the changes in weather, the aircraft detection results may be not so good as we want. What's more, the existing methods need lots of aircraft prior information and need to train large number of samples to achieve good results. On the basis of existing algorithm, this article have done lots of researches and proposed a novel unsupervised aircraft detection algorithm. The main work is as follows:In this paper, a detailed study of existing aircraft detection algorithm has been made, with the overall framework of the most common algorithm based on feature learning summed up. The key technologies of each part of the framework have been analyzed in detail and all the shortcomings have been presented. Then an algorithm based on global appearance filtering has been studied. Moreover, this paper further studied the visual saliency and summarized the frequently-used features and classic models of it, meanwhile, generating the saliency maps of the models and analyzed their applicability in remote sensing images. After that, different saliency maps of the remote sensing image have been analyzed. Finally, this paper applied visual saliency to aircraft detection in remote sensing images. The novel algorithm divided the original problem into two stages as coarse detection and precise detection. Firstly, the circle-frequency filter was utilized to coarsely detect the aircrafts, and then co-saliency was used to precisely detect the real aircraft and eliminate the false alarms. It is an unsupervised aircraft detection algorithm with good experimental results.