Infrared Target Detection Based On AdaBoost With Sea-sky Background

Infrared target detection technology with sea-sky background is an important branch of the target detection area.it is an important technology of target detection with sea-sky background and has become a hot topic with broad prospects for application. AdaBoost is an ensemble method which relies on the principle of generating multiple predictors and weighted voting among the individual weak classifiers.In this thesis, the background and development of infrared target detection with sea-sky background have been introduced, and the theories and techniques of infrared target detection have also been presented. The major task is to apply the AdaBoost algorithm via OPENCV platform into the field of infrared target detection with sea-sky background area, propose a new AdaBoost algorithm for this area and investigate the possibility of applying AdaBoost into this field for further work of exploring real-time and efficient target detection system with the sea-sky background. The main researches are as follows:1. In this thesis, traditional methods of target detecting with sea-sky background have been introduced: infrared image pre-processing, targeting, image segmentation, target detection. Several classical methods of image pre-processing, targeting, target segmentation have been introduced, and simulated the traditional methods of target detecting with sea-sky background.2. The AdaBoost algorithm is applied to infrared target detection with sea-sky background according to the theory and principles of the application of AdaBoost in the area of face-detection. Against the differences of the target with sea-sky background and the target of face, explore the more effective features for sea-sky target. Compared with the feature of SIFT/SURF, the feature of Haar performs a lower time cost and a better performance.3. In this paper, against the high cost of time and the poor performance in detection with the traditional AdaBoost algorithm, an improved AdaBoost algorithm has been proposed and applied to infrared target detection system with the sea-sky background. Against the disadvantage of high time cost and large computation, three methods to improve feature selection have been explored. And against the disadvantage of over-training, sometimes even stuck in the process of training in the case of that it is in lack of samples or the situation that the diversity of negative sample is not enough in the traditional AdaBoost, propose a new weight updating method. And against the disadvantage of bad performance in detection, a new threshold selection method is been explored to ensure the output more objective.