Study On The Identification Of Grape Leaves Based On Manifold Learning

With the development of the grape-wine industry,grape varieties' identification becomes significantly critical for the popularisation and market promotion of this type of cash crop.Commonly,In the study of grape recognition,the leaves were used as the research object,but there was a significant difficulty in the identification of grape leaves.Leaf color,morphological structure difference is small,which caused great obstacles to the identification of the cultivars.In order to help the researchers better solve the problem of identifying grape varieties,the following research has been conducted:In this study,15 types of mature grape leaves were chosen as the research object.For feature extraction,multiple research methods were applied to extract the feature of the grape leaves,including the Grey Level Co-occurrence Matrix,Histogram of Oriented Gridients,Deformable Parts Model feature and Convolution Neural Network.Analysis of the characteristics of the characteristics of data and performance,It is found that the high-dimensional feature indicates that the grape leaves are higher than the low-dimensional features,but the high-dimensional feature data is large and the redundancy is high.Although it can get better recognition results,but the efficiency is low.In order to reduce the complexity of the grapevine recognition model with high dimensional characteristics and improve its practicability,the manifold learning algorithm is used to reduce the dimension of the extracted high dimensional grape leaves.On the basis of maintaining the recognition precision,the efficiency of the algorithm is improved.In the study of feature reduction dimension,we use four different algorithms,such as Locality linear embedding(LLE),Laplacian embedding(LE),Locality preserving projections(LPP)and Neighborhood preserving embedding(NPE),are applied respectively,get the low-dimensional feature of our grape leaves,and the key parameters that affect the dimensionality reduction are analyzed.In the process of grape leaf recognition,the classification effect of different classifiers is analyzed.Then training support vector machine(SVM)Model for leaf classification and identification.In this study,we analysis the feasibility and necessity of manifold descending.The reduced dimension of manifold can effectively keep the internal structure of the data in high dimensional space.After the dimension reduction,the recognition speed is improved,and the accuracy of the recognition is still good before the dimensionality reduction.It is found that the feature extracted by convolutional neural network use the manifold learning algorithm to reduce the dimension,the recognition rate can reach 90.33%,the recognition performance is better than that undo the dimension reduction,and the recognition speed is greatly improved,and the recognition time is reduce to 1/3.For the dimensionality reduction of the artificial design feature,the recognition time are obviously improved.After the DPM feature reduce to 1/30,the recognition time is reduce to 1/6.The study of this paper provides an effective method for the rapid identification of grape leaves...