基于改进的NCSPSO-AFSA对SVM参数的优化及其应用

Parameter optimization and application of SVM based on improved NCSPSO and AFSA

为了找到支持向量机(SVM)最佳的分类参数,用以构建适合纹理图像分割的SVM分类器,文中是将基于小生境和交叉选择算子的粒子群算法(NCSPSO)引入变异算子和族外竞争机制加以改进后与人工鱼群算法(AFSA)混合,提出了一种改进的NCSPSO-AFSA混合算法优化支持向量机参数,并分别与AFSA算法,粒子群算法(PSO),NCSPSO算法在图像分割准确率、参数寻优时间、图像分割时间等方面进行对比和分析,实验表明文中算法能够更好地获得适用于纹理图像分割的SVM参数,在缩短图像分割时间的同时提高了图像分割准确率,相比较其他算法,文中算法稳健性更好.将此方法应用于电镜及超声纹理图像分割中能较好地提取出目标区域,图像边缘部分的分类也很清晰.

To find the best parameters of SVM and construct the SVM Classifier which is suitable to be applied to texture image segmentation , the paper improves niche and cross-selection operator PSO where mutation mecha-nism and group competition mechanism are introduced and combined with artificial fish-swarm algorithm （ AF-SA） .The parameter optimization algorithm for support vector machine （ SVM） is proposed .Compared with AF-SA, particle swarm optimization and NCSPSO algorithm in accuracy and time of image segmentation and parame -ter optimization time , it turns out that the proposed algorithm can effectively find the parameters of SVM , cut time and improve accuracy of image segmentation , at the same time , its stability is better than other algorithms . When applied to the electron microscope image and ultrasonic image segmentation , the method it can extract the target area and the image edges are classified quite clearly .