【摘要】 为实现机器视觉准确判别葡萄干品种,提出了一种基于压缩感知理论(Compressed Sensing,CS)的葡萄干品种分类方法。以3种葡萄干为研究对象,并提取葡萄干图像的形态、颜色和纹理特征参数,得到葡萄干训练样本的数据词典矩阵。压缩感知理论分类算法首先利用由葡萄干图像特征参数组成的数据词典矩阵对每一个葡萄干测试样本进行稀疏性表示,得到稀疏向量。然后利用稀疏向量对葡萄干测试样本进行重构,并计算重构样本与测试样本之间的残差,最后通过比较残差的大小来确定测试样本的类别。将提出的方法与最小二乘法支持向量机(Least squares support vector machine,LSSVM)和BP(Back Propagation)网络的识别结果做了对比和分析。试验结果表明,基于压缩感知理论的分类方法对于3个葡萄干品种的综合分类准确率为99.17%,获得了最好的分类效果。更多还原

【Abstract】 A classification method based on Compressed Sensing was proposed for discriminating the varieties of Raisin precisely. Three kinds of Raisins were investigated,and the shape,color and texture feature parameters of the Raisins image were extracted,and then,the data dictionary matrix was got. Firstly,this classification process was to represent the test samples of Raisins image by the matrix of data dictionary and to obtain the sparse vector. Secondly,the residuals were calculated between the reconstructed samples and the test samples by making use of the sparse vector to reconstruct the test samples of Raisins image. Finally,the classification of test samples were determined by comparing the sizes of residuals. In this study,the classification results on the proposed method were analyzed and compared with those of Least Squares Support Vector Machine( LSSVM) and BP network. Experimental results demonstrated that the overall classification accuracy of Compressed Sensing method is 99. 17%,which has the best classification effect among three methods.更多还原