基于支持向量机的甜柿表面病害识别

Recognition of Persimmon Surface Disease Based on Support Vector Machine

研究以次郎甜柿为研究对象,应用快速独立分量方法对病害图像进行处理,去除病害图像中的随机噪声。提取病害甜柿表面图像的纹理特征参数和颜色特征参数后采用支持向量机(SVM)识别甜柿表面病害。研究表明快速独立分量方(FASTICA)法可以很好的扩展病害区域边缘。采用图像纹理特征参数和颜色特征参数结合构建支持向量时,识别准确率和速度都高于只用单一特征参数的情况。在选择图像纹理特征参数和颜色特征参数结合作为支持向量的基础上,对结构参数进行优化选择,结果表明:当SVM类型为nu-SVR,核函数为Sigmoid,模型特征参数C=26、ξ=24时所建立的识别模型对果体表面病害的识别效果最好。研究为快速无损检测次郎甜柿的病害提供了一定的理论依据和方法。

The image parameters of persimmons disease surface had been extracted by machine vision technology which combined with Fast Independent Correlation Algorithm（Fast-ICA） and Support Vector Machine（SVM） method in order to enhance the identification precision of disease persimmon.First of all,using Fast ICA method to remove the random noise in the persimmon’s disease image.Then,the disease region in image had been segmented by mathematical morphology.Finally,the texture feature parameters and color feature parameters from image of persimmon surface disease had been extracted.Using support vector machine to identify the surface defects persimmon.The confirmatory experiments indicted that： fast ICA method acquired remarkable effect in extension of the persimmon disease region edge.Combined the texture feature parameters and color feature parameters in image of persimmon surface disease as the support,the vector was proved to be higher accuracy and velocity than just using the single image feature parameter.Different parameters were chosen to optimize the support vector machine as well as construct the recognition model on the ground of the conclusion above.The model indicated that when chosen the SVM type as nu-SVR,the nuclear function as Sigmoid,the model characteristic parameter C=26,ξ=24,the best forecast effect could be obtained（the precision is 98.5%）.The model forecast effect parameters RMSEC reached 0.9831,and the RMSEP reached 0.5404.