Research And FPGA Implementation Of Infrared Dim Small Target Image Preprocessing Technology

Infrared guidance head can detect and track target, it has been widely used in surveillance, searching, early warning and imaging guidance due to its working all day capability. Actually we hope the detector can detect target in a far distance as soon as possible, so more precious time can be saved for decision and preparation. However, when the detector is far from the targets, they become dim and small and image signal to clutter ratio is too low, so target detection is difficult. In addition, with the enhancement of target maneuverability and improvement of weapon, people put forward more requirements towards the stability, real-time, and miniaturization of guidance head, so it’s significant for us to do research on high performance infrared dim small target detection technology.In infrared dim small target image processing technology, the performance of image pre-processing will immediately affect the performance of recognition and tracking in next stage, so high performance pre-processing technology is foundation of whole system. Considering that there is less study on city ground scene and multiple small target detection technology is not ready. On the basis of deep learning of related field at home and abroad and combining with requirement of our subject, an infrared dim small target image pre-processing system solution is proposed. In this thesis jobs I finished are as follows:Firstly, the system framework is designed and key technologies are analyzed.Secondly, target, background and noise feature of infrared image are analyzed. Comparison is made for commonly used single frame infrared small target detection methods in ground scene. Due to the shortage of these methods, a multi-scale top-hat background suppression method is proposed, which can suppress background effectively and improve the ability to detect target with different size. Then a target contrast enhancement method is proposed based on pattern lateral inhibition theory, which further improve local signal to clutter ratio of image and provide advantages for latter target extraction.Thirdly, in the latter stage of target recognition, the guidance head may tremble strongly, so the false recognition rate may be raise. An infrared video interframe jitter parameter estimation method is proposed to solve this problem. This method is based on conventional registration technology and it is improved to run in hardware.Lastly, all methods of pre-processing system are implemented in FPGA, including multi-scale top-hat background suppression module, target contrast enhancement module, target adaption extraction module and jitter parameter estimation module.All methods in this thesis are verified by simulation, experimental results show that our methods can detect small targets efficiently when original local signal to clutter ratio is lower than 4, single frame target detection probability can achieve to 95%. The FPGA pre-processing system is also verified, it can meet the 30 frames per second real-time processing requirement.