基于形态学分水岭的垩白米粒检测方法

Detection of chalky rice based on morphology and watershed algorithm

提出了一种基于形态学分水岭的垩白米粒检测方法。充分利用单个垩白米粒图像区域中垩白部分白色不透明这一物理特征,将单个垩白米粒图像区域分成3个标记区域,通过形态学分水岭方法分割垩白米粒图像,提取区域间的单像素边缘将垩白区域提取出来。为了提高边缘检测的准确性,使用sobel算子去噪。由于使用标记控制分割,从而使区域极小值修正具有自适应性和准确性特点。试验结果表明:该方法对垩白米粒的检测精度为96.4%,而且识别效果好于基于分形维数的检测算法。

A detection method of chalky rice based on morphology and watershed algorithm is proposea, which can detect various chalky rice kernels only by edge extraction. The chalky kernels is classified into three regions as normal（or white） region and chalky（or chalky white） and background（or black） region according to their aspect color features. Because the method of morphology and watershed can exactly extract the edge of the different regions by using the sobel algorithm,the extraction of the chalky region of chalky rice kernels can be adaptive. In our experiment, the proposed method is tested on over 2 400 rice kernels with normal kernels and 800 chalky kernels,and the detection accuracy of the chalky kernels is 96.4%.