Research On The Algorithms Of Infrared Dim Small Target Detection

In Infrared(IR) Image Processing area, the technology of Infrared(IR) small target detection has been concerned by scholars for years. Especially in early warning, guidance, remote sensing etc., this technique has been extensively adopted. However, not only because the imaging result of IR dim small target is poor and the target size is relatively small, but also the noise and backgroundclutter wave are strong, the target signal is always submerged in the background. Therefore, for the purpose of advancing the reliability and steadiness of the detecting system, and to detect and track this class of objective, scholars usually focus on the research of background suppression and the detection algorithms in all kinds of mode. Based on the extensive research of the wavelet threshold denoising algorithm, this paper put forward three kinds of wavelet thresholding denoising based IR dim small target detection algorithms, and then put forward a space domain dynamic threshold target confirmation strategy, and finally put forward a dynamic queue based pipeline filter algorithm, and a complete set of IR dim small target detection system. The key contents and results of this paper are as following:(1) According to the study about the detection methods in wavelet domain, a detection algorithm based on the generalized cross validation criterion and wavelet thresholding denoising is raised. The generalized cross validation criterion can effectively estimate wavelet subband threshold and does not need to estimate the noise variance, which can prevent from the error caused by inaccurate noise calculation.(2) A detection algorithm based on Wavelet and activity level estimate is proposed. Adaopting the criterion of minimum mean square error, the activity level of current point is obtained by the weighted average of neighborhood points. Then by estimating the energy of points with similar activity level, the spatially adaptive Bayes threshold can be calculated, thus the denoising thresholding is precise to the pixel dimensions.(3) Based on three wavelet threshold denoising algorithms, the traditional Bayes, GCV and spatially adaptive Bayes thresholdings, this paper proposes a strategy of spatially adaptive multiple model denoising. This strategy selects the most suitable wavelet threshold function according to the different characteristics of wavelet subband noise, which makes the detection results advance a lot.(4) This paper puts forward a multi directional difference factor based spatial domain detection algorithm. This algorithm identifies and detects the targets through the definition of gradient difference function of pixels in each direction of the infrared image. Based on this, this paper further puts forward a dynamic threshold strategy to confirm the target. This strategy slides the threshold of multi difference factor and the parameter of two value segmentation from high to low, in order to make analysis and judgment on the potential targets which are identified in previous algorithms of the system from less to more, thus can confirm targets effectively.(5) This paper raises a dynamic queue based Pipeline filter algorithm. Through the set up of dynamic update queue, it can effectively capture the true targets and eliminate false targets by using the information of inter frames, and finally outlet real-time pipeline targets, which can effectively complete the task of dim small target detection and confirmation.(6) Combined with the wavelet domain denoising method raised in(3), the spatial domain method put forward in(4) and the time domain method presented in(5), this paper designs a system for infrared dim small target detection, realizing the image sequence detection for single target and multiple targets, making the system to be with high detection probability and low false alarm probability and with better stability and real-time.(7) The receiver of characteristic(RO C) curve, which is extraordinary in common use in industry, is introduced in this paper for the IR dim small target detection algorithm evaluation. The curve can measure the denoising ability and background rejection capability of infrared dim small target image of an algorithm by calculating the area between the coordinate axes with the lower half and the curve, thus can reflect the performance of the algorithm indirectly. Finally, this paper makes comparison and analysis of several algorithms by the simulation of single frame and multi frame image sequences, which corroborated the availability and real-time results of the presented detecing methods and system.