Camellia Pest Image Pattern Classification Method Of The Research Based On SVM

The camellia is a very important part of the forest ecosystem in the south of China,it can not only provide high quality edible for people,but also purify the air, beautify the environment, cover the land, protect land and water resources, it has enormous ecological and social benefits.At present, in our country,camellia output of per mu yield is very low,pest is one of the important reasons lead to lower productivity.In this paper, This article took camellia pests as the research object, applying image processing and pattern recognition technologies to achieve camellia pest image identification and classification. This paper mainly completed the following work:(1)In order to reduce the noise in the image, improve image quality, we need to process the image of the collected oil-tea pest image, through analysis and comparison, using histogram equalization enhancement, then using median filter technology for image smoothing.After the image noise decreases obviously,this paper proposes a new method of the maximum difference neighborhood and regional merge segmentation algorithm, so as to realize the oil-tea pest image segmentation.Through the simulation experiments show that compared with the OSTU method, method of the maximum difference neighborhood and regional merge segmentation algorithm can obtain better result.(2)In the feature extraction stage of camellia pests, the paper discusses the characteristics of global features such as color, shape, Invariant moment and focuses on the local features of the Scale-invariant feature transform (SIFT) features, and put forward the BoW(Bag Of words) model references into the field of oil-tea pest image classification,which widely applied to text categorization.This paper expounds that the image recognition are carried out using this model which needs to be done for upfront work process.(3)This paper selected classifier designed by SVM method,Through the experimental simulation, we obtained different kernel function and the input feature vectors of training, testing and the result of classification.The experimental results showed that the method took 83.3% of the average recognition rate, the highest recognition rate reached 87.3%, achieved the intended purpose of oil-tea pest image pattern classification.