Research On Recognition Method Of Plant Leaf Images Based On Manifold Learning Algorithm

Agriculture is the foundation survival conditions of the human resistance, and it is need to understand different crop growth characteristics and physiological and biochemical parameters in order to improve crop yields. Identification and classification accurately is the first premise of monitoring the growing crops. Contrast to crop plants, the leaves may live for a long time and collection conveniently, and there are abundant nutrition and growth information in plants leaves which are benefit to distinguish from others plants. So it is possible that using plants leaves as classification features to recognize the plants category. In recent years, the manifold learning algorithm is added in recognition methods of plants leaves. The method main idea is that using the manifold learning algorithm to reduce the feature dimensions of leaves images which are extracted after preprocessing to cluster the leaves images features information. The existing recognition methods of leaf images in real plant leaf images database have achieved some good classification accuracy, but these are some disadvantages in these methods, such as extracting features, supervised manifold learning algorithm and effectively classifier for classification and so on.According to the disadvantages of the presently methods of the plants leaves images, the article aims to study the features extraction、 manifold dimensional reduction and classifier, proposing some new methods to identify plants leaves. New methods mainly to extract high-dimensional features from leaf images at first, and manifold learning algorithm is used to reduce the high-dimensional features dimensions to attain the low-dimensional embedding; Then in a low dimensional space using the classifier to identify the plant leaves category.; Finally, based on the real leaf images database, the simulation experimental results are analyzed. The main research contents and results are as follows:(1) Putting forward a new method that combines the weighted locally linear embedding (WLLE) to SVM to recognition plants leaves. This method uses the weighted local linear embedding algorithm (WLLE) to reduce high dimensional color features which are composed by all pixels of leaf images, then using support vector machine (SVM) to recognize the category of the leaves in low dimensional space. The method solves the nearest neighbor classifier cannot effectively identify plant leaves category in low dimensional space. The experimental results show that the method proposed improves the recognition accuracy of leaves images.(2) Proposing a recognition method of multi-feature plant leaves based on dissimilarity-supervised locally linear embedding algorithm.The method extracts the features of color、shape and texture as leaves multi-feature at first, and then the sample dissimilarity is brought into weighted locally linear embedding to form the supervised LLE algorithm to reduce leaves multi-feature dimension, at last, the nearest classifier is used to recognize leaves category in low dimension space. The method solves the problems that traditional leaf recognition methods cannot fully extract leaf images characteristics and the existing manifold learning algorithm cannot fully mining sample’s category information. The experimental results based on real plant leaf databases show that the proposed method improves the leaves recognition accuracy effectively.