The Research On Polarization Imaging And Processing Technology Based On Embedded System

Polarized optical imaging technology is a new imaging technology based on the detection of the polarization information of the target's reflected light.Compared with traditional intensity imaging technology,polarized optical imaging technology can reflect differences in water content,roughness,texture characteristics,physical and chemical properties of different targets,and is conducive to highlighting targets that are difficult to identify with traditional imaging technology.In this paper,the polarization imaging and processing technologies are studied from three aspects: polarization information acquisition,polarization information fusion processing,and embedded system implementation.In terms of obtaining polarization information,this paper conducts research on interpolation and parameter calculation based on the division of focal plane polarization imaging technology.Starting from the principle of the division of focal plane polarization imaging technology,the causes of instantaneous field of view error in this imaging method are analyzed in detail.The compensation effect of the interpolation algorithm on the instantaneous field of view error is studied,and an edge-based polarization information interpolation method is proposed to reconstruct the Stokes vector and calculate the polarization feature image.The experimental results show that compared with other interpolation methods,the polarization feature image obtained by the method in this paper is closest to the ideal reference image,and has greatly improved in subjective visual perception and objective evaluation.In terms of polarization information fusion processing,this paper studies the fusion method of polarization feature images,and proposes two methods suitable for polarization feature image fusion.The first method obtains the multi-scale and multi-directional sub-band coefficients of the image by NSCT transform,and then uses the simplified pulse-coupled neural network model with adaptive parameter and adaptive energy weighting of the region to fuse the high-frequency sub-bands coefficients and low-frequency subband coefficients of the two images respectively.The second method uses the guided filter to obtain the basic layer and the detail layer of the image.On the basic layer,the fusion rule combining the pixel energy and the gradient energy is used,and on the detail layer,the fusion rule with the largest absolute value is used.The experimental results show that compared with other polarization image fusion methods,the two polarization image fusion methods proposed in this paper have greatly improved in subjective visual perception and objective evaluation.The first method is more complex and difficult to implement on embedded system;the second method is simple and fast,and is more suitable for implementation on embedded system.In terms of embedded system implementation,this paper selects the Jetson-TX2 development platform produced by NVIDIA as the core device.The edge platform-based polarization information interpolation,polarization parameter calculation,and guided filtering-based polarization image fusion method are used for the embedded system.Among them,the parallel computing scheme of GPU is used for the polarization parameter solution part.Finally,Qt Creator is used to design a graphical user interface and implement the algorithm visual operation.