Research On Lu’an Crispy Peach Quality Grading Based On Computer Vision

The peach planting area of Lu ’an city in Anhui province is about 230,000 mu,ranking first in the province.The early harvest time of crisp peach in Lu ’an is 15 to 20 days earlier than that in other provinces and cities,which has high economic value.Peach cultivation plays an important role in local rural poverty alleviation and rural revitalization.At present,Lu ’an crisp peach is in the transition stage from quantity to quality.In the circulation of commercialization,the quality of crisp peach is an important factor affecting its economic benefit.There are mainly artificial and mechanical classification methods.Their disadvantages are: high input,low efficiency,classification based on the characteristics of disunity and easy to cause mechanical damage,classification characteristics of a single.Thus restricted the development of peach.In this paper,lu ’an crisp early peach as an example,hereinafter referred to as crisp peach.The color,size,shape and skin defect of crisp peach were evaluated by computer vision.The main work is as follows:(1)A set of operable evaluation indexes was developed based on the analysis and study of the local standards of crisp peach and relevant literature.And 200 peaches were selected in Taoyuan,and the research data set of crisp peach classification was made according to the index.(2)Extracting the appearance features of crisp peach.Firstly,the ratio of red and green related pixels on the surface of crisp peach was calculated,and the maturity of crisp peach was divided.Then,the minimum circumferential circle method was used to extract the cross diameter of crisp peach and obtain the size parameters of crisp peach.Finally,the improved roundness value analysis method was used to analyze the shape characteristics of crisp peach.(3)Deep learning method was used to detect peel defects of crisp peach.Four target detection models including Mask R-CNN,Center Net,YOLOv5 and Efficient Det were compared in the experiment.Finally,the Efficient Det algorithm was selected with a good balance between prediction accuracy and prediction speed,and its accuracy reached85.83%.(4)Design a comprehensive grading system software based on Efficient Det algorithm.The results showed that the classification accuracy of crisp peach was 100% superior fruit,97.5% second fruit,96.6% third fruit and 100% equal outer fruit.Compared with the traditional classification method,this classification system has higher recognition accuracy and recognition speed,and avoids mechanical damage.It provides ideas for automatic and intelligent grading of crisp peach.