Research And Implementation Of Camellia Fruit Visual Positioning System

with the progress of science and technology,the system of each part of the picking robot is also increasingly perfect.Among them,the design of robot vision positioning system has a great impact on its work efficiency,especially in the speed of target detection,the accuracy of picking fruit,and the adaptability to the picking target environment.The main content of this study is based on convolution neural network Camellia fruit positioning system.The image of Camellia fruit is obtained by binocular stereo vision equipment,and the depth information is collected and calculated.The VOC data set of Camellia fruit is made.The improved yolov3 target detection algorithm is used to realize the recognition of Camellia fruit in complex environment.Finally,it is found that the improved yolov3 algorithm has higher recognition rate and faster recognition speed,and through the design of the host computer interface,the positioning function of Camellia fruit target is intuitively displayed.Traditional image processing methods mostly use the feature description method to obtain its color and texture features to complete the recognition,and its feature extraction ability is poor in the complex environment.This study uses the feature extraction network based on deep convolution neural network.In the complex natural picking environment,the trained model is easier to obtain the more complete image features of the target,and has higher efficiency Accuracy and robustnessAt present,there are many algorithms of convolutional neural network applied to image,each of which has its own advantages and disadvantages.Through the analysis and comparison of some classical convolutional neural network algorithm structure,according to the purpose of this study is to detect the Camellia fruit target,this paper analyzes and compares R-CNN(region convolution neural network),Fast R-CNN(fast region CNN),YOLO(you only look)in detail Finally,it is decided to improve the adaptability based on yolov3 algorithm.Aiming at the goal of Camellia fruit,this paper puts forward three improvement methods: the improvement based on network depth,the improvement based on spatial pyramid pooling layer,and the improvement based on network training strategy.The algorithm complexity test proves the advantages of the improved yolov3 algorithm.In the process of Camellia fruit target collection,in order to improve the robustness of the system and ensure the integrity and diversity of samples,this study uses binocular vision equipment to collect the target sample information of Camellia fruit under different time,angle and other environmental factors,and uses label Img image annotation tool to label the data set to establish the VOC data of Camellia fruit Set.In the experimental design of Camellia fruit detection,the accuracy,recall and recognition speed of the improved algorithm,the original algorithm and RCNN series algorithm are compared and analyzed.Experimental results show that the improved yolov3 algorithm has higher recognition efficiency,and shows its superiority in complex environment.