【摘要】 以玉米田间杂草图像为研究对象,对采集的杂草叶片图像进行预处理,对图像的多个纹理特征进行筛选,以支持向量机进行分类。针对传统分类器的不足,以组合核函数对其性能进行优化。仿真结果表明,构建优化的组合核函数能使分类器性能得到显著提升,且当组合核函数中径向基函数所占的权重为0.2、多项式核函数(二阶)所占的权重为0.8时识别率最高,达86.00%,可以满足杂草识别的需求。更多还原

【Abstract】 Image preprocessing of corn field weed blade was used to extract texture characteristic. The support vector machine(SVM) was used for classification. Considering the shortcomings of the traditional classifier, the parameters were optimized via the composite kernel. The results of simulation showed that the new method improved the weed identification. The highest recognition rate of test samples was 86% when the weight of radial basis function was 0.8 and polynomial kernel function was 0.2.更多还原