Research On Disease Spot Grading Of Huangguan Pear Appearance Detection Based On Machine Vision

China is a major producer and consumer of pears in the world,with pear cultivation area and yield ranking first in the world.Huangguan pear is one of the most common pear varieties in China.Due to the diversity of individual shapes and lesion types of Huangguan Pear,traditional computer vision techniques and pattern recognition methods have certain limitations in detecting disease spots in Huangguan Pear.Currently,it almost entirely relies on manual testing,which has low cost,efficiency,poor timeliness,and strong subjectivity.In recent years,with the rapid development of computer technology and machine vision technology,it has been widely used in crop disease detection,providing a new solution for rapid and accurate detection and grading of Huangguan pear disease spots.Therefore,this paper takes Huangguan Pear as the research object.Firstly,feature extraction methods are used to extract the appearance parameters of Huangguan Pear,and then focus on the deep learning based Huangguan Pear disease spot detection and classification algorithm.The main research contents of this paper are as follows:(1)This paper uses the Canny operator to extract the appearance features of Huangguan pear.Collect and process images of Huangguan pear by constructing a hardware platform for machine vision system.Using Canny operator to extract the edges of Huangguan pear and obtain the number of pixels.The lateral and longitudinal diameters of Huangguan pear are obtained through the contour search algorithm,with accuracy rates of 97%and 97.71%,respectively.The mass and volume are obtained by multiplying the pixel point information and the correction coefficient,with accuracy rates of 94.92%and 97.51%,respectively.This provides a reference for the rapid detection of appearance and weight of Huangguan pear.(2)This paper proposes a Huangguan pear grading method based on instance segmentation,semantic segmentation,and hierarchical network fusion.Firstly,a Mask RCNN based method for complex background segmentation of Huangguan pear was proposed.Improve the CLAHE image enhancement module to CLAHE-Mask R-CNN and conduct comparative experiments with Mask R-CNN.The experimental results show that CLAHEMask R-CNN has more advantages in segmentation accuracy,with a PA of 97.38%.Due to the subjectivity of manual grading of Huangguan pear fruit,a semantic segmentation based Huangguan pear disease spot segmentation model was proposed.The current mainstream semantic segmentation models DeepLab V3+,UNet,and PspNet were used for segmentation training and comparative experiments were conducted.The experimental results showed that the PA of DeepLab V3+reached 92.87%,the Dice coefficient reached 67.25%,and the IoU coefficient reached 84.36%.In order to further improve accuracy,DeepLab V3+has been improved by adding an ECANet attention mechanism module to the original DeepLab V3+,which has increased by 0.65%,1.60%,and 2.48%compared to the original models PA,Dice,and IoU,respectively.Finally,three mainstream grading models ResNet50,MobileNetV3,and VGG16 were used to grade Huangguan pear.The Huangguan pear image,which was segmented using CLAHE-Mask R-CNN and subjected to improved DeepLab V3+calculation of lesion ratio,was used as a dataset input for training in the network.The use of this dataset can avoid the interference of human subjectivity on the grading of Huangguan pear,provide an accurate dataset for the grading model,and improve the reliability of the model.By using a test set to test the model,RseNet50 achieved an average accuracy of 96.67%.This method provides a theoretical basis for the fruit grading of Huangguan pear.(3)This paper constructs a classification method for Huangguan pear in complex background environments based on segmentation and recognition of Huangguan pear,disease spot segmentation of Huangguan pear,and grading of Huangguan pear.This model framework integrates the results of three stages of the model to achieve the classification of Huangguan Pear in complex background environments,effectively reducing the impact on Huangguan Pear classification in complex environments.