Visual Recognition And Maturity Detection Technology Of Guava In Natural Environment

With the rapid development of computer and automatic control technology,the level of agricultural informatization and intelligence has been greatly developed.However,the current agricultural production still mainly depends on mannual work or agricultural machinery controlled by manual,and the degree of automation needs to be further improved.Machine vision technology has the characteristics of high recognition accuracy,rich information,non-destructive testing,etc.It can provide technical support for fruit and vegetable growth monitoring,yield estimation,visual positioning of picking robot and quality grading,etc.It has great application value and economic benefits.In this paper,guava is taken as the research object,and the research on visual recognition and maturity detection technology of guava in natural environment is carried out.The main contents and conclusions are as follows:(1)The recognition and localization research on guava in natural environment was carried out.In this paper,YOLOv3 network with higher detection accuracy and faster detection efficiency is selected as the identification method of guava.Firstly,a large number of image sample data were collected and labeled artificially to produce a data set for training YOLOv3.Then,anchor box,batch,initial learning rate and other parameters were determined through experiments.Finally,YOLOv3 network was trained with the parameters determined by experiments.The average accuracy of the final network in the verification set is 88.4%,the recall rate in the test set is 92.7%,and the recognition accuracy is 94.8%.It has high monitoring accuracy and provides technical support for guava visual detection.(2)A classification study of guava maturity was conducted.In this paper,two maturity classification methods were used,based on color feature and SVM and based on improved Le Net-5 network.Using the color characteristics of guava as a classification feature,the SVM classifier was used for classification,and the classification accuracy rate on the test set is 90.9%.Using the improved Le Net-5 network as the guava maturity classification method,the classification accuracy rate on the test set is 94.7%.The improved Le Net-5 network has a higher classification accuracy,which is finally used to classify guava maturity.(3)Guava recognition and maturity classification algorithm were integrated.In this paper,the guava recognition method based on YOLOv3 and the guava maturity classification method based on improved Le Net-5 network are integrated.The integrated algorithm accurately identifies guava and the correct classification rate is 91.1%.The average detection time for one image is 0.13 s,which provides technical support for the identification and maturity classification of guava.In this paper,using the machine learning technology YOLOv3 and the improved Le Net-5 network,the visual recognition and maturity detection of guava in the natural environment is realized.The accuracy of accurate identification and classification is 91.1%,and an average image is required for image detection.The time is 0.13 s,which can provide visual technical support for the estimation of the production of guava orchard and the positioning detection of the picking machine.At the same time,it also has reference and reference value for the identification and maturity detection of other fruits.