Research On Mobile Measurement Method Of The Number Of Tomato Fruits With Different Maturity Based On Color Point Cloud Image

Accurate evaluation of number of tomato fruits with different maturity on tomato plants is not only helpful to fine-tuning the amount of fertilizer applied in the process of tomato planting,but also helpful to make plans for tomato fruits harvest,storage and sales in advance.The growth forms of tomato plants are different,and the color of immature tomato fruits is similar to that of leaves,and tomato fruits are prone to overlap and be covered.Complex environmental background makes it difficult to recognize tomato fruits.Under the complex environment background,accurate identification of tomato fruits is the key to accurately evaluate the number of tomato fruits with different maturity.At present,domestic and foreign scholars mainly focus on the two-dimensional image data of single view as the research object of crop fruits recognition,but the two-dimensional image data of single view can not reflect the distance information of crop fruits,and can not avoid the problems of crop fruits overlap and occlusion.In order to solve these problems,this study designed a mobile measurement method of the number of tomato fruits with different maturity based on color point cloud image,in order to measure the number of tomato fruits with different maturity in greenhouse.The main research contents and conclusions are as follows:(1)A mobile measurement system of the number of tomato fruits with different maturity was built.Aiming at the information of light environment and spatial structure of tomato plant distribution in greenhouse,a mobile robot platform with the ability of autonomous mapping navigation and high-precision movement was selected,and data acquisition equipment was equipped to form a measurement system of the number of tomato fruits with different maturity,which can realize fixed-point docking and collect image data information of tomato plants.(2)The single view color point cloud image data of tomato plants were obtained.This study analyzed and compared different point cloud information acquisition methods,and selected Kinect V2.0 camera based on time of flight(TOF)ranging method as the single view color point cloud image data acquisition device.According to the camera parameters obtained by calibration,the RGB image collected by Kinect V2.0 camera was mapped to the depth image to get the RGB-D image,and the RGB-D image was converted to the color point cloud image through the three-dimensional coordinate transformation formula.(3)Two view tomato plant point cloud image registration was realized.In order to solve the problems of tomato fruit occlusion and overlap in single view tomato plant point cloud,a point cloud image composed of two view point cloud images with different positions was proposed as the data processing object.Different point cloud filtering methods were analyzed and compared,and threshold method and statistical filtering method were selected to preprocess the point cloud data of single view tomato plant.This paper analyzed different point cloud registration methods,combined random sampling consistency algorithm and feature matching algorithm for coarse point cloud registration,and used the improved ICP registration method based on point to plane for fine point cloud registration to obtain the final composite point cloud image data.(4)The fruit recognition of tomato plant image data was realized.This paper analyzed two kinds of object detection methods based on traditional computer vision and deep learning,established Point RCNN object detection network based on deep learning to recognize tomato fruit in tomato plant point cloud,and took the tomato fruit recognition method based on super pixel feature vector based on traditional computer vision as contrast method.Taking the artificial statistical value as the reference value,the accuracy rate of tomato fruit recognition by pointrcnn target detection network is 86.19%,the recall rate is 83.39%,and the average center relative error is 36.83%.The accuracy rate of tomato fruit recognition by contrast method using two-dimensional image data as the data processing object is 83%,and the recall rate is 75.8%.(5)The classification of tomato fruits with different maturity was realized.According to the internal color distribution of tomato fruit,tomato fruit can be divided into four stages:Green Maturity,Slight Maturity,Maturity and Complete Maturity.Different color features of tomato fruit were extracted to form a color feature matrix,and the color feature matrix was used to establish a classification model to judge the maturity of tomato fruit based on the support vector machine(SVM)classifier.The prediction accuracy of the classification model for the training set and the test set was 94.27% and 96.09%,respectively.The classification model was used to classify the maturity of the identified tomato fruits.