Research On Detection Of Small Weak Infrared Target Based On Multi-feature Fusion In Background Of Sea-sky

With the development of photoelectron technique, computer technique and detector technique, more and more applications are based on the automatic target recognition (ATR) technique。The task of infrared target detection is an important and difficult issue in automatic target recognition system. The module of target detection is in the front of ATR system whose performance mostly depends on the target detection. The weak target with size of several pixels or dozens of pixels embedded in complicated background has no obvious geometrical structure and grayscale distribution. The disturbance of clutters makes the target detection even harder. This dissertation deeply studies fusion detection for small weak target in background of sea and sky, and presents the fusion detection technique based on multi-feature and multi-level fusion strategy in time and space region. The main contributions of this thesis are given below.The framework of small target detection based on multi-frame and multi-feature in time and space region is proposed. Adaptive fusion detection method and distance map method based on multi-feature are presented under the advanced framework。The complicate background determines the method, which only depends on one frame and one feature, can not detect small weak target. These features are utilized, such as local maximum gray level feature, morphologic feature and local entropy feature. Adaptive fusion detection method determines the weights of features automatically based on peak value and average gray scale of background in feature images. And the weights vary with different image sequences. Multi-feature distance map method detects small targets based on distance measure and fusion scheme in multi-feature vector space between each feature vector and the average feature vector of the line.Experimental results show that the multi-feature fusion methods reduce the probability of false alarms (Pf) and improve the signal to noise and probability of target detection (Pd). The presented methods have high performance.