Research On Key Technologies Of Underground Target Feature Extraction And Imaging Based On Ground Penetrating Radar

GPR feature extraction and imaging technology is currently widely used in military,national defense,people’s livelihood and other fields.With the development of signal compression and reconstruction technology,GPR imaging based on compressed sensing technology and target feature information extraction based on depth learning become the main force of GPR signal processing.At present,there are many experimental data of BP algorithm in GPR imaging algorithm,and the speed is slow and the accuracy is low.This algorithm is integrated with wavelet algorithm and compressed sensing technology,and the signal reconstruction technology in compressed sensing is improved to form a set of radar imaging methods with high sampling efficiency and high imaging accuracy.By combining this imaging method with the You Only Look Once v5(YOLO v5)target detection algorithm,a set of high-precision and fast calculation algorithms for ground penetrating radar imaging and target feature information extraction were obtained.The specific research content is as follows:1.A fusion algorithm of wavelet denoising and Back Projection(BP)imaging is proposed to address the problem of large amounts of noise in the signal received by the radar receiving antenna.Due to the limitations of the hardware of the radar transceiver antenna,the electromagnetic wave signal received by the radar system contains a large amount of noise.By integrating the data sampling algorithm with the wavelet algorithm,the signal is denoised before the imaging process,further improving the imaging accuracy of the ground penetrating radar.Through simulation experiments,the imaging accuracy can be increased by 1.63 times.2.Aiming at the problem of algorithm operation speed in wavelet denoising and BP imaging fusion algorithm,an improved technique of compressed sensing reconstruction algorithm is proposed.This thesis analyzes the characteristics of GPR imaging,uses compressed sensing technology to sample data in the process of processing GPR echo signals,and proposes an improved reconstruction algorithm based on conjugate gradient method in view of the characteristics of the Orthogonal Matching Pursuit(OMP)reconstruction algorithm in the compressed sensing reconstruction algorithm,such as low reconstruction accuracy and large iterative residual error.This method can effectively improve the reconstruction speed.According to simulation verification,this reconstruction method can greatly improve the algorithm speed while ensuring that the imaging accuracy changes little.In the algorithm,wavelet de-noising technology,improved compressed sensing technology and BP imaging algorithm are fused,and the final fusion algorithm can improve its imaging accuracy by 1.60 times,with a speed of 32.3s.3.A YOLO v5 network target feature extraction algorithm based on SE net is proposed for higher requirements of ground penetrating radar imaging recognition accuracy.By adding the SE net module to the YOLO v5 network,we aim to enhance the network’s utilization of target feature channel signals.Extract target feature information from the processed image using an improved algorithm.This target feature information extraction method can extract underground target feature information from ground penetrating radar images with high efficiency and accuracy,and compared to the original YOLO v5,the network m AP has increased by 3.4% and the Precision has increased by 4.8%.