Research On Weak Target Recognition Technology Based On Sparse Aperture Imaging

With the widespread use of various small UAVs in civil and military fields,various threats and challenges are posed by them.Therefore,the effective detection and identification of small and medium-sized UAVs in low-altitude areas has become one of the hot spots in the research of target identification.Small UAVs are characterized by small size and complex flight environment,and it is difficult to detect such low-altitude,slow and weak targets by traditional technical means,not to mention their accurate identification and localization.Sparse aperture optical imaging system adopts multi-aperture structure design,which can not only reduce the volume and weight of the optical imaging system,but also make it achieve the effect of high-resolution imaging.The sparse aperture optical imaging system is applied to the detection of low-altitude slow and weak targets such as UAVs and combined with deep learning target detection algorithm,which has the characteristics of small size of detection system,long imaging distance,high resolution imaging and can quickly and accurately identify and locate targets,etc.,and has great prospects for application in the field of ground-to-air high-resolution imaging detection.In this paper,we focus on the weak target recognition technology after sparse aperture imaging,and the main research contents are as follows:(1)The MTF of a typical sparse aperture array is simulated and the imaging characteristics of the sparse aperture system are analyzed.A sparse aperture imaging degradation model is constructed,and the imaging is verified by simulation simulations for typical sparse aperture arrays.The image recovery preprocessing experiments were conducted with the Wiener filter,Richardson-Lucy,and constrained least-squares filtering algorithms for ring-type sparse seven-aperture images,respectively.Based on the subjective and objective evaluation results of the image preprocessing experiments,it is known that the Richardson-Lucy algorithm has a good image restoration effect on the annular sparse seven-aperture image data.(2)The characteristics of sparse aperture imaging targets and low-altitude slow weak targets are analyzed,and the key issues of weak target recognition technology based on sparse aperture imaging are studied.By analyzing the current typical target recognition algorithms and considering the accuracy and recognition rate of target recognition,the YOLOv5 algorithm of single-stage target detection is selected as the weak target recognition algorithm of sparse aperture imaging in this paper.For the characteristics of weak targets in sparse aperture imaging,four methods such as data enhancement algorithm Mosaic-9,4-fold feature extractor,CBAM attention module and EIo U bounding box loss function are used to make targeted improvements to the YOLOv5 algorithm.(3)A ring-type sparse seven-aperture imaging system was built to perform imaging acquisition experiments on a small UAV flying at low altitude and slow speed.The acquired image data were preprocessed using the Richardson-Lucy image recovery algorithm,and then the processed image data were labeled with targets to produce the sparse aperture image dataset SA-UAV.training comparison experiments were conducted for the original YOLOv5 algorithm,the YOLOv5 algorithm with four improved strategies and the improved full fusion YOLOv5 algorithm.According to the experimental results,the m AP value of the improved YOLOv5 algorithm reaches 97.7%,which is 8.3% higher than the original algorithm,and the target recognition rate in the sparse aperture imaging video reaches an average of 17 ms per image,indicating that the improved target recognition algorithm can simultaneously take into account the recognition accuracy and recognition rate for the weak targets of sparse aperture imaging.