Research On Denoising And Target Detection Of Ground Penetrating Radar Image

Ground penetrating radar technology has the advantages of high resolution,high efficiency,intuitive radar images and no damage to buried objects,so it is widely used in the field of underground buried objects detection.However,because the image is inevitably interfered by different noises during the acquisition process,and these noises seriously affect the quality of the buried object reflection waveform in the geological radar image.So making the detection accuracy of the buried object to decrease.For this problem,although the traditional geological radar image denoising technology can filter out some of the noise in the image,it will destroy the edge details and texture features of the target in the image to varying degrees,thereby reducing the detection performance of the target in the image.Resulting in false alarms or missed alarms in engineering applications.Therefore,it is urgent to study a denoising method that not only has good denoising performance but also can protect the edge detail texture characteristics of the target body from being damaged.(1)A joint denoising algorithm combining bilateral filtering and block-matching and 3D(BM3D)algorithm is proposed,which can effectively remove the noise in the ground penetrating radar image and keep the detailed texture features of the target edge from being destroyed.The joint denoising algorithm makes full use of the advantages of bilateral filtering to preserve edge denoising and the BM3 D algorithm to filter out Gaussian white noise and retain image details and texture features.The algorithm flow chart is given,and two performance indicators(peak signal-to-noise ratio,structural similarity)that characterize the denoising effect of geological radar images are compared and analyzed.The experimental results show that with the increase of the peak signal-to-noise ratio,the denoising effect of the proposed algorithm is better than that of the denoising method when the spatial domain,frequency domain and spatial frequency domain are used alone.Especially when the signal-to-noise ratio is small,it not only has higher peak signal-to-noise ratio and structural similarity have also been significantly improved,so that the image can achieve better denoising performance while effectively retaining the edge details and texture features of the target volume.(2)In order to improve the detection and classification accuracy of buried objects in ground penetrating radar images,an improved Faster RCNN network model is constructed based on faster region convolutional neural network(Faster RCNN).Firstly,this article collected the measured data sets of different buried objects and the simulated data sets output by GPRMax3.0 under the same configuration environment through field experiments.Then use the improved Faster RCNN target detection framework and use VGG16 as the feature extraction network of the buried object detection and classification model,next use the K-means clustering algorithm to improve the generation size of the region suggestion box.Then the model is trained and tested supervised by actual data set and simulated data set.The experimental results show that the m AP value of the measured multi-target image under the proposed method is increased by about 2.06% compared with the Faster RCNN model,and its best grade can reach 92.56%.Meanwhile,the m AP value of the simulated single target and the measured single target are improved about 0.48% and 1.14% respectively.Therefore,the proposed improved network not only has better detection and classification effects for single-target and multi-target simulation images and measured images,but also has strong robustness for the detection and classification of measured multi-targets.