现有煤矸分选主要有人工分选和机械分选，这些方式存在劳动强度大、能耗高、易造成环境污染等问题。近年来，煤矸分选机器人的研究受到业内广泛关注。对煤矸分选机器人而言，煤矸的准确识别是一个关键且具有较大难度的问题。研究了基于图像的煤矸识别方法，并在此基础上开发了识别系统。介绍了煤矸分选机器人中图像识别系统的硬件组成，研究了实际工况条件下各部件的选择和安装方式;在实验室搭建图像采集系统，选取韩城矿区的煤和矸石为样本，由所搭建的系统获取样本图像，建立了样本图像库;对样本图像采用3种不同的滤波器进行降噪处理，对比分析得出非线性低通滤波处理效果最佳;基于煤和矸石表面物理特性在灰度和纹理两方面有一定的区别，分别对煤和矸石样本图像的4个灰度参数和5个纹理参数进行分析对比，得出在灰度方面灰度均值和最大频数对应的灰度值2个参数区分度更高，在纹理方面纹理对比度和熵2个参数区分度更高;选用最小二乘支持向量机(LS-SVM)为煤和矸石图像识别分类器，以灰度均值和最大频数对应的灰度值组成的灰度特征、纹理对比度和熵组成的纹理特征、最大频数对应的灰度值和纹理对比度组成的联合特征作为分类器的输入向量分别对分类器进行训练和对比验证，得到以联合特征进行训练的分类器识别效果更好;以LABVIEW为平台开发了包括图像采集、图像滤波、联合特征向量的提取、样本分类等程序。在煤矸分选机器人实验平台上搭建了识别系统，随机选取实际工况下的煤和矸石样本，对识别系统分类性能进行测试，系统图像降噪采用非线性低通滤波器，分类采用联合特征训练的分类器。测试结果显示煤和矸石分类准确率分别为903%和83.0%,平均识别时间为0.153 s。

Currently,the sorting of coal and gangue mainly relies on manual sorting and mechanical sorting. These two methods are labor intensive,consume a large amount of energy,and cause environmental pollution. Therefore,in recent years,the research on coal and gangue sorting robots has drawn much attention. One of the key functions of the sorting robot is identifying coal and gangue,however,which still remains a crucial and difficult problem to be solved. This pa- per propose an image processing based method for the problem and further develops an identification system. The hardware composition of the system in the coal and gangue sorting robot is introduced,especially,the selection and installa- tion methods of the components of image identification system under the real-world condition are studied. Firstly a coal and gangue image repository is constructed by building image collection system and collecting the images of coals and gangues from Hancheng mining area. Then,an experiment is conducted to compare three kinds of filters for noise re- duction of the images,which indicates that the nonlinear low pass filtering achieves the best performance. Considering that the surfaces of coal and gangue differentiate in grayscale and texture,they are compared in terms of four parame- ters of grayscale and five parameters of texture, it is found that the coal and gangue are more distinct in the two grayscale parameters including gray average and the grayscale value corresponding to the maximum frequency,and oth- er two texture parameters including contrast and entropy than other parameters. Furthermore,LS-SVM is chosen as the image classifier. With the training of the classifier by inputting the two grayscale features,the two texture features and the combined features of grayscale and texture respectively,it is found that the classifier using the combined features has the best performance. The programs have been developed for the image collection,image filtering,combined feature vector extraction,and sample classification using LABVIEW. The identification system is built on the sorting robot ex- perimental platform. To evaluate the performance of the system,the images of coals and gangues are chosen,which are randomly picked from production environment. Furthermore,the nonlinear low pass filter is used for noise reduction and the combined features are used to train the classifier. The results show that the model achieves an accuracy of 90． 3% in identifying coals and 83% in identifying gangues,the averaged identification time is 0. 153 s.