Research On IR Small Target Detection And Tracking Based On Attention Mechanism

As key techniques in infrared (IR) homing guidance, target search and tracking, warning and so on, IR small target detection and tracking have been regarded as old-line and attractive research topics in the field of IR image processing. As for the real weapon systems, how to make the best of IR target detection techniques to increase the distance of target detection and to obtain the related information about the invading targets, have become important factors to decide the victory or defeat of modern warfare. The longer the distance of target, the less the imaging area of target and the larger the probabilities of targets influenced by backgrounds and clutter will be. Therefore, comparing with other topics in the field of IR target detection and tracking, how small targets can be robustly detected and tracked under complex backgrounds have become the more realistic and challenging research topics.Detecting and tracking algorithm of moving dim small targets in IR images with complex background are investigated in this dissertation. The main work can be summarized on image preprocessing, target detection and tracking.Aim at the problem that different size and shape have a large effect on the result of morphological filtering, a novel method for self adaptive morphological Top-hat operator in background suppressing of small target was presented, and the structural elements of the operator are optimized by advanced Genetic Algorithm (GA), adaptive updating strategy was used to control the GA crossover rate and mutation rate and the niche technique based on the method of maintaining optimum is adopted in the GA training step, which reduces the possibility of premature convergence presence and improves the exploitation capabilities of GA.Comparing with the traditional algorithms, the experimental results show that the proposed algorithm can preserve the detail image to the greatest extent, reduce the influence on the background estimation, improve the signal-to-ratio (SNR) greatly and the detection probability in single frame. Traditional algorithms such as "detect before track (DBT)" and "track before detect (TBD)" are studied, which need compute all region of the image to judge whether the target exists in the sight field, even though the target occupy a small region. An infrared small target detection algorithm based on visual attention mechanism is proposed in this paper to solve this problem. An infrared small target image is divided into inside and outside scene, to the inside scene, a method based on minimum error probability (MSE) is applied to extract the Region of intrest (ROI); to the outside scene, a method based on multi-feasure fusion is applied to identify the targets. The visual attention-based approach reduces the computation complexity, while the other performance aspects are not traded off. The experimental results indicate that the method can effectively detect multi-targets in low signal noise rate infrared image sequences, especially for the realtime detction in the large sight field.Based on the algorithm of visual attention mechanism, a high performance infrared small target detection system based on TMS320C6416s is designed and implemented in this dissertation, In this system, the data processing units are three pieces of TMS320C6416, loose couple and strar-like structure is adopted. The system is designed under modular designing idea based on DSP and FPGA. An application example on tracking infrared small target under complex background indicates that this system has good reconfiguration, real-time ability and applicability. The experimental results show that the precision of angle measurement and real-time performance can meet the requirement of design index.Based on the analysis of the cause of sample impoverishment, quanta genetic algorithm was introduced into the particle filter (QGAPF) to solve the problem. Sample impoverishment was relieved by increasing the diversity of samples set, and the ability of estimation and tracking were ameliorated. Experimental results demonstrate that the proposed algorithm can alleviate the effect of the sample impoverishment phenomenon for the particle filter. It is applied to the real infrared small target tracking and the obtained results are compared with particle filtering (PF) and Extended Kalman Filter (EKF). Experimental results show that QGAPF has advantages in the field of state estimation problem.In summary, the infared target detection and tracking problems are researched in this paper, and new algorithms have been proposted.