Large Depth Of Field Anti-scattering Imaging And Positioning Method Based On Deep Learnin

In the real world,the existence of complex media,such as air molecules,dust particles,haze and water vapor,prevents people from directly obtaining ideal scene images and distance information,which has a serious impact on the safety of driving in foggy days.Therefore,studying how to obtain information,such as the structure and distance of the targets through the scattering medium,is of great significance to many fields,such as people’s livelihood,military affairs,and scientific research.Based on the basic principles of optical scattering,combined with speckle related imaging model and phase space measurement theory,a datadriven anti-scatter imaging and locating technology is proposed in this paper with large depth of field,and two anti-scatter cooperative reconstruction and perception algorithms are designed,to achieve high-quality reconstruction and high-precision positioning of the scattered targets.This paper carries out the following work from two aspects,image reconstruction and depth localization of the original targets:(1)In terms of image reconstruction,in view of the poor image reconstruction quality and unstable imaging results under the large depth-of-field scattering scenes,an anti-scatter imaging algorithm is designed in this paper based on a U-shaped structure,combined with scatteringrelated imaging methods,to improve the algorithm’s extraction of targets features ability.The anti-scatter imaging network is used to reconstruct targets of different depths,sizes and plane positions in a large depth of field.Experimental results show that the PSNR of the designed algorithm can reach 22.8670 dB in the 1150 mm large depth of field scattering scene,the recovery results are still stable when the depth and size of the targets to be measured change at the same time,and it has better imaging generalization ability under the condition of small stride displacement of the target.(2)In terms of depth locating,in order to solve the problem of poor targets positioning accuracy in large depth of field scattering scenarios,an anti-scatter locating algorithm is proposed in this paper based on deep learning,combined with the principle of phase space depth resolution,and introduces a frequency domain conversion module to effectively improve the convergence efficiency of the algorithm.The positive effect of DTF layer on network efficiency and locating accuracy was verified by comparative experiments on anti-scattering locating.The depth locating of speckles based on high-precision rail acquisition is carried out,and the accuracy error at 1150 mm under the condition of large depth of field is only 0.4823mm;The depth resolution ability of the locating network in the small step displacement and the target size location change is analyzed,and the resolution error is less than 0.4mm;The locating generalization ability of the network under the condition of small stride is analyzed,and the accuracy error is only 0.2465mm;Finally,two reconstructed perception synergy networks with different performances are designed to meet the efficiency and accuracy requirements of different scenarios.This paper a large depth of field anti-scatter imaging and locating method is designed in based on the physical prior model,combines the data-driven thinking to mining and extract the speckle image features from the two aspects: image reconstruction and targets locating,improves the accuracy and stability of the output results,designs two collaborative reconstruction perception networks with different performance focuses,improves the applicability of anti-scattering imaging positioning technology to different application scenarios.