Blind Restoration And Information Extraction Of UAV Remote Sensing Blurry Image

With the rapid development of remote sensing technology,micro-UAV remote sensing platform has gradually become an important way to obtain crop information in precision agriculture because of its high efficiency,small size and simple operation.It is also the main way to achieve crop yield estimation,crop type identification and pest control.However,due to the light weight of UAV,it is easy to be affected by atmospheric disturbance during flight,and there are human operation errors and other interferences,resulting in the blurry phenomenon of crop images obtained during operation,which has a serious impact on the subsequent crop recognition,information extraction,pest recognition and other research work.In this study,the blind image restoration algorithm is used to restore the millet image obtained by UAV,and the research of information extraction is carried out.The specific research contents and conclusions are as follows:(1)The experimental area is located in Zhongyang village millet planting base,Taigu District,Jinzhong City,Shanxi Province.The image was collected on September 27,2019,from 16:00 p.m.to 18:00p.m.at this time,millet is in the filling period.The phantom4 advanced UAV produced by Dajiang science and Technology Co.,Ltd.is equipped with FC6310 visible light camera with 5474 pixels × 3648 pixels to obtain blurred images of millet crops during operation.The weather condition on the day of image acquisition is good.In addition,in order to accurately extract crop information and estimate the yield of millet spikes,the UAV was designed to navigate at an altitude of 4m and fly at a speed of 3m/s.In this experiment,a total of 18 blurred crop images were collected.In order to speed up the image restoration process and better show the restoration effect,each image was divided into 6 image blocks,and a total of108 image blocks were obtained.(2)Three representative blind restoration algorithms in recent years,namely,the blind restoration algorithm based on normalized sparsity,the blind restoration algorithm based on dark channel prior and the blind restoration algorithm based on learning discriminant fitting function,are used to restore the blurry crop blocks obtained by UAV operation.On this basis,a novel image reweighted graph total is proposed The main principle of the reweighted graph total variation(RGTV)is to restore the edge weight of the blurry crop image to the bi-modal distribution of the original clear image,and remove the noise contained in the crop image,so as to get a more accurate blur kernel;hyper Laplacian priori is to use the estimated blur kernel to quickly achieve non blind image restoration,and finally get a clear restored image.The restoration results of the three popular algorithms and the proposed algorithm are measured by subjective visual evaluation and objective evaluation index values.In this process,in order to better measure the effect of each algorithm,the initial blur kernel of each algorithm is set to the same size.The experimental results show that compared with the existing deblurring algorithms,the smoothing effect of the restored image obtained by the proposed algorithm is significantly weakened and contains less artifacts;and in most cases,its objective evaluation indexes,namely Information entropy,Standard deviation and Mean value,are higher than those corresponding to the other three algorithms,which is conducive to extracting crop related information from the restored crop images.The Entropy,Standard deviation and Mean value obtained by the algorithm are 2.3%,3.3% and 2.1% higher than the original blurry images respectively.Compared with the three algorithms,the maximum value of the three evaluation indexes is increased by 0.28%,8.6% and2.08%,which is helpful to extract the crop related information from the restored image.(3)When extracting the crop information from the restored millet images,it is observed that the color characteristic of the ear and the senile leaf is similar at the late filling stage and the ever-maturing stage,which will cause some interference to the crop information acquisition using the agricultural UAV equipped with the visible light camera.Therefore,morphological algorithm is firstly used to preprocess the millet spike and leaf from the restored image,and then extract six common shape and texture feature parameters with RST invariance(the rotation,scale and translation of the image are invariable),including the roundness and texture of the ear and leaf.The texture feature parameters include the extraction of energy,information entropy,contrast and homogeneity of millet spike and leaf.Then SVM classifier is used to select the common kernel function to realize the classification and recognition of millet spike and leaf.The experimental results show that the polynomial kernel function is the most suitable for the classification and recognition of millet spike and leaf,and the average recognition rate of the two is 94.83%.Then,the yield of millet was estimated based on the experience of millet planting,and the accuracy rate reached 85.93%,so as to achieve the purpose of accurate extraction and classification recognition of UAV blurry crop image information.The restoration algorithm proposed in this study can effectively predict the blur kernel,suppress the ringing effect,and better restore the edge details of crop images,so as to achieve more accurate crop information extraction.This research method provides theoretical support for the use of UAV equipped with visible light camera to achieve more accurate low altitude remote sensing crop information monitoring,and has certain reference significance.