Automatic Plant Leaves Classification Based On Leaf Shape And Vein Structure

It has been more mature for plant leaf classification combined with image processing, pattern recognition technology. But the commonly classification method is few, and classification feature is single, and category overly depends on people. This research automatically classifies 210 kinds of leaves by extracting shape and vein features. The main work is as follows:The first part, the significance of this research and the research status of such studies at home and abroad are introduced. To reflect the whole classification process is automatic, the petioles’ removing uses morphological processing method in the image preprocessing. Leaf image graying, binarization, morphological processing have been done in image preprocessing for preparing for the feature extracting.The second part, extraction and analysis of the shape and texture features of plant leaf image. First, extract contours and minimum bounding rectangle of leaf images preprocessed, then calculate shape features that the paper needs, such as squareness, elongation, equivalent circle radius, circularity, ellipticity, eccentricity and seven Hu Invariant Moments. Then get the GLCM of leaf images which already have been graying, and get energy entropy of leaf images and so on by GLCM, then calculate the texture features. Finally, the experimental analysis of the effectiveness of shape and texture features.The third part, vein features’ extraction of leaf images has been done in this part. Using a modified Sobel operator detects 8 directions of vein, then fusion the eight directions picture of vein to obtain a complete vein image; and then preprocess the vein image, such as Venus filtering, vertical and horizontal directions’ dilation, median filtering, thinning and so on. Finally, extract the veins bifurcation points and end points, and calculate the distance of each point to the center point(point of maximum curvature vein) to get vein characteristics. Eventually, it is the experimental analysis of the effectiveness of the vein characteristics.The fourth part, the classification effect of the support vector machine classifier compares with the BP neural network using the same shape and texture features which are in this paper. Select the classifier this research uses-Support Vector Machine. Then classify the testing leaf images by the support vector machine using the shape and texture features to finish the initial classification. The correct accuracy is 91%. Then reclassify the un-completely classified leaf images using the initial features and vein features by SVM. The reclassified correct is 96.1%.