Research Of Plant Leaves Classification Based On Image Processing And SVM

Plant classification and identification is the basis and premise of the Plant Protection andResearch, it has far-reaching significance for the division of plant species and to explore theevolution and phylogenetic relationships of the plants. The shape of the characteristics ofplant leaves to the classification and identification of the plants is the most simple andeffective method. There are low efficiency,large randomness and other drawbacks for thetraditional artificial identification method. In this paper, five kinds of shape features of fourkinds of plant leaves：papaya buergerianum, Acer mono, privet was extracted for each120photos of the total480photo, then use the support vector machines(SVM)modeling tocomplete the identification and classification. On this basis, combined with MATLAB GUItools to build a plant leaf identification system.The main work and conclusions of the paper isas follows:(1) Discussed image acquisition and pre-processing methods. The typical shape leaveswere captured by images with a digital camera, then the leaves image pre-processing methodswere analyzed, including the gray-scale processing, denoising, edge processing,morphological filtering, background segmentation, identified effective image preprocessingmethods.(2) Researched leaf feature extraction and efficient features choose method. Leaf areaand perimeter parameters obtained by the binary image connected component labeling, on thebasis of the extracted rectangular degree, elongation, roundness, density and invariantmoments these five moments shape characteristics and entropy, energy, the same quality,contrast, correlation these five texture parameters. These characteristic parameters hasinvariance characters of translation, scaling, mirroring and rotation invariance, which haspractical significance of the classification.(3) Studied a combination models and methods of particle swarm optimization and SVMfor leaf classification. Shape features, texture parameters were combined with particle swarmparameter optimization algorithm to build a support vector machine (SVM) classificationmodel for plant leaves classification and ultimately leaves the accurate classification resultsshow that the recognition rate is is98.86%. However, the average recognition rate of BPneural network is86.9%while Fisher Discriminant method of average recognition rate is 85.75%. A comparative analysis with BP neural network and Fisher discriminant method wasmade, shows that the method we proposed was better than the other.(4) based on support vector machines (SVM) recognition algorithm proposed in thispaper leaves, using MATLAB's GUI toolkit design a plant leaves recognition system whichprovide a commercially viable ways and means for the automatic identification of plantleaves.