Based On Optical Remote Sensing Image Target Detection Algorithm

Target detection in remote-sensing image is very important both in military application and civilian use. Target detection has strong pertinence for types of target and images, so target detection algorithms must be carefully designed in order to have good results. In this paper, target detection algorithms in remote-sensing image are researched, aiming to detecting bridges target and aircrafts target. In addition, a shore-ship separation method used to identify ships on the dock is researched which is based on segmentation result of the port remote-sensing image. An improved watershed segmentation method is specially researched to segment the port remote-sensing image which solved the over-segmentation problem that the traditional watershed segmentation has. Main works of this thesis are summarized as follows:(1) Describe relatively four popular watershed transformation currently. Compare their time complexity through simulation, and draw to a conclusion that the chain code watershed had the lowest time complexity, which laid a foundation for fast regional segmentation based on watershed transformation.(2) Study and realize an improved regional segmentation algorithm based on watershed transformation for port remote-sensing image. This algorithm overcomes over-segmentation of the traditional watershed segmentation. Firstly, the image is filtered by mean filter and noisy is reduced, which reduced over-segmentation; secondly, floating-point image instead of gradient image is used to watershed transformation, and fatherly reduced over-segmentation. Watershed transformation result is merged based on area and average gray value criterion. Improved segmentation algorithm makes effective segmentation for port remote-sensing image.(3) Study and realize a shore-ship separation algorithm. This algorithm is realized through Hough transformation and features of the coastline. Firstly, polygon vertices of water region are calculated, and then the binary image of real water region is available through regional filling. The water region is available through the fusion of the binary result and original image, which solves the identification of docked ships. Currently, similar literatures are little. (4) Study and realize an aircraft detection algorithm based on template match. Aircrafts of different direction can be matched through spinning the template, which is the improvement of algorithm. This algorithm solves a problem that aircrafts of different direction would produce shadow of different direction, which makes the aircraft having a different direction from template not be detected well.(5) Study and realize an effective bridge detection algorithm. An improvement of this method is that it makes a good segmentation. Color and texture feature of river segment are both used to extract the river segment. Bridges are detected by combination of binary morphology and the feature of bridge across the two sides of the river.(6) A remote-sensing image segmentation and object detection system is designed and realized. The system realizes the basic function of human-computer interaction, and new algorithm can be added.