Recognition And Measurement Of Phenotypic Traits Of Cucumber Fruits And Leaves Based On Machine Learning

The cucumber is one of the main vegetables in China,which is widely planted and very popular with people.Its appearance quality influences the market value for consumers.Moreover,phenotypic traits of cucumber fruits and leaves play an important role in breeding.Cultivation of cucumbers that are straight and have short carpopodium,non-wart and good resistance to powdery mildew are meaningful for researchers.However,conventional methodology for manually phenotypes screening is not only timeconsuming but also subjective.Therefore,it is extremely urgent to seek an effective and accurate solution.In this project,an image acquisition system is designed for cucumber fruits and leaves based on machine vison.Taking the particularity of cucumber traits into consideration,this paper focuses on the study of the automatic grading method of fruits,the grading of powdery mildew and warty recognition.For the difficulty of extracting phenotypic traits of cucumber fruits and leaves in image processing,several key algorithms such as segmentation of cucumber seed cavity,segmentation of leaf scab and warty recognition are proposed based on machine learning.Finally,an automatic extraction platform for phenotypic traits of cucumber fruits and leaves is developed.Following the national standard,an automatic grading method of fruits is proposed.The method extracts cucumber contour by image filtering,shadow elimination,binarization and other operations.Then the melon length,diameter,diameter difference,pedicel length,arcuate height of cucumber fruits are computed through skeletonization algorithm and interpolation fitting algorithm.In the study of pedicel length measurement,a method,based on the random forest algorithm,is proposed to solve the problem of cucumber seed cavity segmentation,which combines color and texture.The grading of cucumber leaf powdery mildew is deeply studied and a method to automaticaly identify leaf scab is proposed.After the analysis of the sample images,the Z channel in the XYZ color space with large contrast is selected as the classification feature.Then,combining the fuzzy C-mean algorithm and the simple linear iterative cluster algorithm,the leaf scab is identified effectively.Finally,the experimental results show the method is superior to the other three algorithms.To measure warty density accurately,a method of warty recognition is proposed in this study.According to convolutional networks,a model is built based on U-net architecture,which can be used for image segmentation after automatic extraction of relevant features.The experimental results show that the method can improve the accuracy,compared with the traditional segmentation method.The basic version of automatic extraction platform for phenotypic traits of cucumber fruits and leaves is designed based on the image acquisition system and the above algorithms.To meet the need of the promotion,a portable version is developed in this thesis.The experimental results show that the method has high efficiency and high precision,which presents an effective solution for cultivation of new cucumber varieties.