Research On Defective Peaches Recognition And Orientation Based On Binocular Vision

With the improvement of people’s living standards,the demand for fruits is also increasing.And peaches are known as the king of fruits due to their crispy taste and health benefits.During the growth and subsequent market circulation of peaches,various skin defects such as decay,cracking,and insect bites may occur,which greatly damages the appearance of peaches and seriously reduces their economic value.Traditional peach defect detection mostly relies on human eyes,while defect detection technology is mostly based on traditional image processing and uses invisible light for detection,which requires high requirements for the detection environment and equipment.With the development of deep learning technology,defect detection technology based on deep learning has been increasingly applied.This article will take common peaches as the research object,complete the detection of defective peaches based on deep learning,and locate and grasp the target defective peaches to promote the automation development of the peach processing industry.The main work content of the paper is as follows:Firstly,the collection of peach images was completed,and defective peaches were classified according to national standards.Corresponding image preprocessing was carried out to address quality issues in the images,and the dataset was expanded to improve its scale.Finally,image annotation was performed to complete the construction of the peach defect dataset.Perform structural analysis on the Yolov4 algorithm,use peach defect dataset for pre training,obtain detection results,and determine the direction of network improvement based on the phenomenon of poor detection performance for small-scale defects.Improve the feature fusion module and attention mechanism module of the model,and construct an improved Yolov4 network structure.Finally,design experiments to verify the effectiveness and superiority of the improved model.The internal and external parameters of the binocular camera are calibrated using Zhang’s method and the MATLAB camera calibration toolbox,obtaining the camera’s internal and external parameter matrix,and completing distortion correction based on the calibration results.Perform stereo correction on the left and right camera images,and then calculate the peach centroid and radius based on Hough circle transformation.Establish and use a competition function to find the corresponding positions of the centroid points in the left camera image in the right camera image to complete stereo matching,and complete the depth value solution.Build an experimental platform and conduct grab experiments.The experimental platform mainly consists of a binocular camera and a robotic arm.Firstly,the conversion relationship matrix between the camera coordinate system and the robotic arm end coordinate system is obtained through hand eye calibration.Combined with the previous camera calibration results and depth value calculation results,the conversion from pixel coordinates to the robotic arm base coordinate system is completed.Finally,complete the grab experiment and analyze and summarize the experimental results.