Research On Plant Image Reconstruction Method Based On Compressive Sensing

With the rapid development of agricultural informatization and automation,agricultural internet of things,agricultural robots,automatic identification of diseases and insect pests and monitoring of crop growth state are inseparable from agricultural plant image big data.In order to improve the transmission efficiency and relieve the storage pressure,a fast and effective compression and reconstruction method is needed for the mass plant images.Because of its high sampling rate,the traditional Shannon sampling theorem generates a large amount of redundant data in the acquisition of plant images,which is likely to cause data redundancy and serious waste of resources.The compressive sensing theory overcomes the defects of the traditional sampling theorem,realizes the high-precision reconstruction of the original image signal through a small number of observations,and relieves the pressure of storage and transmission.In this paper,the compressive sensing theory is applied to the collection and reconstruction of plant images,and an in-depth study is carried out on how to realize the rapid and accurate reconstruction of plant images and the compression and reconstruction of color images.The main research contents are as follows:(1)Aiming at the problems of low reconstruction accuracy and long reconstruction time existing in traditional compressive sensing methods,regularized adaptive compressed sampling matching pursuit based on Dog-Leg is proposed.Kinect 2.0 is used to obtain the color image of the target plant.The color feature of HSV space and the contour feature extracted by Sobel operator are used to input into the Itti model to construct the saliency feature map,which simplifies the complex background.On the basis of CoSaMP algorithm,regularization is used to ensure the accuracy of support set selection in the iteration process.Combining variable step size adaptive and Dog-Leg least squares algorithm,the limitations of unknown sparsity are broken and reconstruction efficiency is improved.The experimental results show that compared with CoSaMP algorithm,the PSNR of plant saliency feature map is increased by 5.19 dB on average and the reconstruction time is reduced by 6.24 seconds in the range of sampling rate from 0.40 to 0.60.This algorithm can achieve fast and accurate reconstruction.(2)Aiming at the problems of noise and background interference in the plant images collected by Kinect 2.0,the depth data of plant images are preprocessed by using TOF characteristics of Kinect 2.0 sensor,and K-means double clustering algorithm is used to extract the effective region of plant.At the same time,considering the local characteristics and structural characteristics of the signal,and combining Dog-Leg least square method for iterative optimization,a compressive sensing reconstruction algorithm based on non-convex low-rank optimization is proposed.The experimental results show that the reconstruction performance of the proposed algorithm is better than that of the NLR-CS algorithm in both noise-free and white gaussian noise environments.In addition,when the SNR of the measured value is within the range of 15~35dB,the PSNR of image increases by 1.12 dB on average.With the increase of SNR,the PSNR increases by a higher level and has stronger robustness.(3)In this paper,the reconstruction of color images is studied.The traditional joint sparse model is improved according to the correlation between the three channels of color images R,G and B.And combining with the above image preprocessing and non-convex low rank optimization algorithms,a low-rank optimal compressed sensing reconstruction algorithm of the improved sparse joint model is proposed.The experimental results show that TV,BM3D-CS,NLR-CS and the proposed algorithm can reconstruct the color image after sparse transformation using the improved sparse joint model.When the sampling rate is 0.30,the image reconstruction time is 7.97 s,the PSNR reaches 41.07 dB,and the structural similarity is 0.9571.All of them are better than the other three contrast algorithms and refactoring is more efficient and accurate.