基于PSO优化SVM的纹理图像分割

TEXTURE IMAGE SEGMENTATION BASED ON PSO OPTIMISING SVM

针对传统纹理图像分割方法运行时间长,分割准确率较低,提出基于粒子群优化算法(PSO)优化支持向量机(SVM)的纹理图像分割方法。首先在自适应调整惯性权重λ的控制策略中加入PSO中的当前迭代次数和种群数,改进PSO的惯性权重λ的性能;接着运用PSO寻找最优惩罚系数C和高斯核函数中参数γ,然后运用SVM方法对训练样本综合训练建立最佳分类模型,并对纹理图像分割测试。结果表明:对比传统方法,该方法不仅缩短运行时间,分割准确率也得到了提高。同时,对比传统惯性权重对分割结果的影响,改进后的方法使得平均收敛代数减少,寻优时间缩短。

Because of the low segmentation accuracy and long running time of the traditional texture image segmentation method,we present a texture image segmentation method which is based on using PSO to optimise SVM. First,the times of current iteration and the numbers of population in PSO are added to the control strategy of adaptive adjustment inertia weight λ to improve the performance of inertia weight λ of PSO; Then the PSO is employed to search the best penalty coefficient C and the parameter γ of Gauss kernel function; Finally the SVM is used to comprehensively train training sample and to build best classification model,as well as to segment and test texture images. Results show that compared with traditional methods,the new method shortens the running time and makes the segmentation accuracy improved. Meanwhile,compared with the effect on segmentation results by traditional inertia weight,the improved inertia weight reduces the average convergence algebra and shortens the running time.