Research On A Kinect-based Extraction Method Of Apple Tree Canopy Characteristics Information

The canopy is the section of the fruit tree that is exposed to light and abiotic environment,and as the leaves are located in this section,is the main place responsible for photosynthesis and respiration on the fruit trees.The structural characteristics of a fruit trees canopy not only shapes the appearance of different plant shapes but also reflects the growth status and yield potential of the fruit tree.In the modern fruit industry,the establishment of a comprehensive,accurate and automated system capable of information perception,processing,analysis and output of the canopy characteristic configuration providing scientific data support for the development of the new fruit industry.In order to obtain the morphological structure and fruit growth information of apple tree canopy conveniently and efficiently an apple tree canopy information extraction method based on a Kinect sensor and MATLAB software was designed in this study.This method was able to achieve the information perception of the original image of the apple tree canopy,establish the point cloud,the input of characteristic parameters and extraction of feature parameters.The main contents of research and conclusions obtained in this study are as follows:(1)The overall technical scheme of apple tree canopy information extraction based on the Kinect sensor was designed.The hardware and software architecture of the experiment was determined according to three-dimensional structural features of the apple tree.Three different tree structures were identified to determine the three different perspectives.,These three perspectives,which were flat angle,elevation and top face angle.were selected to collect image information based on the stereoscopic characteristics of the apple trees,subsequently achieving information extraction of apple tree canopy characteristic parameters.(2)The 3D point cloud model of the apple tree was established,and the output model underwent image optimization and information processing.The Kinect sensor was used to acquire the original image information of the target fruit tree,the parameters acquired included Color Source,Depth,Color Depth and Infrared Source.Simultaneously,the corresponding two-dimensional coordinates and depth information were obtained,thereafter the 3D point cloud model output of the target object was achieved through the three-dimensional coordinate conversion formula in MATLAB.The noise of the output model is filtered through Gaussian filter and the RGB color in the original image is copied so as to get the primal 3D point cloud model of a single fruit tree with no background and no noise.(3)Different characteristic parameters of apple tree canopy and their corresponding extraction methods for each parameter were studied.The maximum and minimum values of some coordinate points on the canopy were determined by permutation method.In addition,the minimum external moment of the apple tree canopy was established,and its length and height were output so as to obtain the width and height of the canopy.The canopy density was obtained by using the point cloud amount on the canopy to divide by the outer moment volume.Additionally,the identification,segmentation and counting of the fruit in the canopy could be realized by color threshold and the ROI point cloud output methods.(4)The apple tree canopy information extraction experiment based on Kinect sensor was carried out,and the experimental results shows that this method can extract the canopy feature information accurately.According to the results,the width and height information of the canopy extracted in the straight-ahead perspective had the smallest the relative error of 2.91%and 3.68%respectively.Branches and leaves outside the outline of the canopy the main factors which could influence the extraction effect.In addition,the relative error of the fruit information extracted in the head-up view was the smallest and was 20.97%.Leaf occlusion was the main factor that affected the extraction of fruit recognition information.In addition,the density of the canopy could be well described by utilizing the amount of point cloud comparison to the outer moment volume.The densities of the three different shaped tree canopies were 6.15 per dm3,9.675 per dm3 and 11.44 per dm3 which was similar to the actual situation.This method was able to achieve the expected objective of canopy information extraction from apple trees,in addition,it is simple,offers quick analysis and has a low error rate.Therefore,it can provide the technical foundation and decision-making basis for the development of modern fruit industry.