Research On Recognition And Location Technology Of Tomato Picking Target Based On Binocular Vision

In recent years,with the development of artificial intelligence technology,the application research of fruit picking robots has been changing rapidly.As an important core content of picking robots,machine vision has an important influence on the accuracy of target detection and fruit positioning.In this paper,picking tomatoes in greenhouses is the research object,focusing on the research of fruit recognition and spatial positioning under the background of complex growth conditions of tomatoes,and building an Android monitoring platform to achieve real-time detection.The main tasks completed in this paper are as follows:(1)This paper takes the tomato fruits in six growth states,including single,cluster,forward light,backlight,overlap,and occlusion under greenhouse environment,as the research object,and proposes the target detection method of YOLOv4+ migration learning network model,using Image Net data set and VGG The network model initializes the network parameters,and transfers the initialized feature vector extraction module to the YOLOv4 network model to train tomato fruit model weights,and obtain the optimal weights to detect tomato fruits in different growth states.Experimental results show that the average detection accuracy of this model is better than YOLOv4 and YOLOv4-Tiny network models.(2)Establish the Efficient Det network model with Tensor Flow as the framework and Bi FPN as the feature extraction network to realize tomato fruit target detection,and compare the recognition effect with the three network models of YOLOv4+ migration learning,SSD,and Faster R-CNN.The experimental results show that the recognition accuracy of the YOLOv4+ migration learning model is slightly higher than the Faster R-CNN and SSD network models,and is similar to the recognition accuracy of the Efficient Det network model.The average detection accuracy of the YOLOv4+ migration learning network model in the six growth states reached 89.07%,92.82%,92.48%,93.39%,93.20%,93.11%,and the comprehensive evaluation index F1 score value was 83%,85.33%,84.67%,89.33%,85.33%,87%.(3)Research on tomato fruit positioning based on binocular stereo vision.First,the image is corrected on the same plane through the internal and external parameters of the binocular camera,the image target is enhanced by YUV histogram equalization,and the SURF stereo matching algorithm is used to determine the picking center point,Locate its three-dimensional space position,and provide data support for the picking robot’s grasping.The experimental results show that the matching accuracy of the six growth states is 80%,85.45%,90.20%,86.73%,76.92%,71.67%,and the three-dimensional positioning error is15.29%.(4)Using Android development technology to build a fruit recognition system,real-time monitoring and recognition of tomato fruits through the mobile phone can accurately detect the ripeness of the fruit.