Infrared Weak Small Target Detection Algorithm Research

Making use of infrared image to realize the automatic target examination, recognition and tracking is the main development direction in equipments of modern military weapons. The dim targets detection in infrared images is one of the key techniques in military weapon system and it has been the subject of intense investigation in these years. Because infrared sensor is easily affected by atmosphere hot radiation, long distance and sensor noise, the detected targets in infrared images often present like dim targets and drowned in noise. The basic problem inherent to extent the detection range is the detection of small, low observable, moving targets in images and complex background. In order to improve the SNR of images and suppress the background clutter, in this paper, we have analyzed the charactering of small target, noise and background clutter. After that, we recommend some preprocessing algorithms and compare the effect of them. Each way has its own advantage and disadvantage, you can choose the right one by the actual scene and the expectation.As to local neighborhood in an image, the distribution of the value of all pixels will be influenced obviously when small target appears. The entropy of this image block will change a lot from the background block, so we can use it to detect dim small target. When this method is implemented, we will face the problem of choosing threshold after calculating all the entropy of image blocks. To tack with this problem, we use high-pass filter to sharpen target and increase the scattering number of these pixels of target. Then, we search the right block whose neighborhood entropy is the minimum of all the image blocks. Because of the large calculation times, we use genetic algorithm to optimize the method. Though we cannot gain the best solution, we can magnify the size of each block in some sort to detect the image block in which the target appears.Because it's hard to ensure the detection with only one frame, we also have studied the algorithm with image sequences in this paper. Multistage hypothesis test algorithm is used to prune a tree-structured list of candidate trajectory segment at each pixel, the trajectory contain an object is accepted in the end. We studied the algorithm and propose that we may use all the statistic of candidate target pixels' moving direction to forecast the moving direction of target, so as to reduce the amount of candidate target pixels between two frame images. In addition, we find that the MHT algorithm has a firm restrict of moving speed. On the basis of MHT algorithm, we use speed matching to resolve this problem, extending the feasible bound of MHT algorithm.