Research Of The Recognition Method Of Flower Leaves Diseases Based On Computer Vision

As greenhouses used more and more widely, the kinds of flowers of greenhouse has become increasingly different, and the kinds of flower planting area is increasing. Due to the unreasonable chemical fertilizer or some other combination of unpredictable factors, the resulting disease more and more, for the production of our country flowers and plants had a great effect. The price of pest eradication is mainly through the application of pesticides, once greenhouse managers aware of the suspected infected flowers, most of them are based on artificial methods(depending on the color and texture features to identify plant diseases) to identify plant diseases and insect pests,but this method doesn’t have a fixed standard, and the human visual system does not have the objectivity, those traits have led to the error rate of artificial recognition is relatively high, under the situation of blind, many greenhouse managers abuse the amounts of pesticides, caused the crop and soil pesticide residues and so on a series of questions. To make rational use of pesticides, we first need to correctly understand the types of diseases, only correctly identify flowers infected by the disease, can effectively use of pesticides, and ensure the good growth of flowers.With the advances the progress of synthesis technology in artificial intelligence, digital image processing technology and pattern recognition technology, flower leaves disease recognition based on image processing is studied, so as to the reasonable application of pesticides and ensure the flowers healthy growth, and then increase the production of flower, to study this subject has a bright application prospect.In this paper, through the analysis and comparison research of both domestic and foreign, and there are orchid disease leaves as the research object,then for the key technology of the flower leaves disease image preprocessing, disease image segmentation, disease image feature extraction and disease image recognition to carry out a detailed study.1)A preprocessing algorithm for the image of orchid diseases was studied.Through the contrast test, it is concluded that the image after median filtering, noise reduction will get the best pretreatment effect.2)The orchid disease image segmentation algorithm is reseached. As large noise of the flower leaf disease spot area on the edge, and easily affected by flower texture feature, through comparison and experiments of several methods, select the a component of lab color space model using the Otsu threshold segmentation method for orchid leaves disease image segmentation, and can effectively isolate the disease spot, lay a good foundation for subsequent extraction of characteristic parameters.3)Then the feature extraction algorithm of orchid disease image is studied.The boundary of the disease spot is segmented, extract the shape characteristics of the disease spot parameters; Using color moment, spot color feature parameters are calculated; Using gray level co-occurrence matrix, extract the texture feature parameters of disease spot, finally determined the 16 characteristic parameters as the input of the characteristics of the subsequent image recognition.4) Finally the suitable for automatic classification of orchid leaves disease pattern recognition algorithm is studied.Through analysis flowers disease characteristics in this paper, this paper chose SVM classifier which has a very good classification effect on small sample image recognition, but the parameters of the SVM model is difficult to determine, so using of genetic algorithm to optimize the SVM, to obtain the optimal SVM parameters, eventually it achieved the good effect of classification, the average recognition accuracy of the optimized SVM is 87.5%.