Leaf Margin Classification Based On Local Images Feature

Classification of plants has great significance for plants protection, distinguish species and explorethe genetic relationship between plants.Leaf can be used as an important basis for the classification ofplants and leaf margin is an important feature of leaf. So classification of plant leaf margins canachieved the purpose of classification of plant. Plant leaf as an experimental material is very easy toobtain and also has little tissue damage to plants, so it is the most simple and effective way for theclassification of plant.Traditional detection and identification of plant leaf margin is mainly based on Detection Agent, ituses the Detection Agent estimate the number of corners and jags on the leaf margin and it does notspecifically realize the classification of leaf margin. To achieve the classification of leaf margin, thispaper study characteristics of leaf margins, and explore effective ways of plant leaf margin for thepurpose of classification. We proposed a new description of plant leaf margin characteristics. Thismethod uses the ratio of Euclidean distance and inner distance between sampled points to express thepartial punch nature of leaf margin, and this can achieve the classification of the entire, toothed edge,wave edge, leaf margin lobed leaves four categories edge classification. The main work and conclusionsfor the thesis as follow:(1) Image collection and pre-processing. Select the typical leaf margin and complete structure leafimage as experimental material, and then preprocess the experimental material.(2) Studied the method of leaf margin feature extraction and proposed using the ratio of Euclideandistance and inner distance between sampled points as the leaf margin characteristics, then discussed theproposed method of leaf margin classification, analysis the rationality and feasibility of the method,confirm the proposed new feature data can effectively distinguish among different leaf margins.(3) Compared the common classification among neural network, Fisher discriminance and supportvector machine (SVM), select SVM as classification algorithm and chose the kernel functionparameters for better result, using Leave-One-Out(LOO) Cross Validation has been to95.15%accuracyrate.(4) Discussed and analyzed the impact of sampling density when extracted the leaf margin feature.The conclusions can be obtained through experiments: Sampling density affects the validity of feature data. Lower sampling density resulted lower recognition rate, when the sampling density reaches acertain value, classification recognition rate will be stabilized. So it can indicate the effectiveness of theleaf margin classification.