Study Of Near Range Short-wave Infrared Imaging

In order to improve the current evolving form of information war,in order to improve the combat efficiency of guidance weapons,we must improve the intelligent high-precision target recognition and precision guidance capabilities.For ground target near-range weapon system,using fast scanning speed and can transmit frame by frame array camera data acquisition,and according to the characteristics of the typical ground target in the data set,to improve the algorithm model,with deep learning based target recognition algorithm instead of the traditional image processing method,experimental results show that the improved Yolov5 algorithm on the ground target detection effect,better for shielding target and local target recognition effect are improved.The main research contents and work of the paper are as follows:Firstly,through the study of infrared detection principle and infrared radiation propagation model,analyze the function relationship,determine the hardware equipment parameters,build the wire array near-infrared experiment platform and conduct splicing imaging,study the image acquisition principle of the wire array camera,and calculate the scientific image acquisition method.Then,different image filtering denoising and image enhancement algorithms are compared to determine the image preprocessing algorithm suitable for the case of this paper.Secondly,the Yolov5 network is improved according to the characteristics of the typical ground.Change the candidate box for the dataset used in this paper;replace the original most neighbor value for upsampling with transposed convolution,to improve the expression ability of the target feature in the target feature map,to reduce the loss of the target feature information,and to effectively capture the global image recognition rate of the occlusion target.Finally,the adaptive NMS is used to better identify the occluded target.The proposed algorithm is validated on the ground target dataset,compared with various typical deep learning algorithms,and analyzed.The improved algorithm can improve the ground target detection performance,with the m AP reaching 86.4% and the detection speed reaching 34.2FPS.The proposed method can meet the real-time requirements while ensuring the detection accuracy.