Accurate classification and recognition of coal flotation froth is one of the key technologies for intelligent coal separation. At present, the coal flotation process relies on artificial recognition of froth features for adjusting the reagent dosage, which cannot realize the optimal control of the quality of the clean coal product and the cost of the reagents. Therefore, in this paper, it is proposed a method of froth image clustering with feature semi-supervision through selection and label information. It is mainly divided into two stages: offline clustering and online recognition. The offline stage is to preprocess the froth image under various reagent conditions, extract the morphology, colour and texture features, and select the multi-dimensional optimal froth image features. A small number of marked samples are introduced to optimize the Gaussian mixture model. The selected optimal features are integrated into the optimized Gaussian mixture model to construct a froth image clusterer with multi-dimensional optimal features and class labels. In the online stage, the real-time froth image features are input clusterer and compared with the cluster feature samples to identify the current reagents conditions, which is used as feedback information to guide the abnormal reagent conditions during the production process. The effect of the amount of supervision information and the quality of feature on clustering results is analyzed and compared through experiments. The application results show that this method can provide key technical support for the accurate control of the dosage of reagents and the quality of clean coal product in the coal flotation production process, reduce the cost of reagents and the number of production accidents, improve the economic benefits, and promote the development of coal flotation intelligence to a higher level.

浮选泡沫的准确分类与识别是智能选煤的关键技术之一。目前，煤浮选工艺依靠人工识别泡沫特征来调节试剂用量，无法实现对洁净煤产品质量和试剂成本的最佳控制。因此，本文提出了一种通过选择和标签信息对特征半监督的泡沫图像聚类的方法。它主要分为两个阶段：离线聚类和在线识别。离线阶段是在各种试剂条件下预处理泡沫图像，提取形态，颜色和纹理特征，然后选择多维最佳泡沫图像特征。引入少量标记样本以优化高斯混合模型。将选定的最佳特征集成到优化的高斯混合模型中，以构建具有多维最佳特征和类别标签的泡沫图像聚类器。在在线阶段，将实时泡沫图像特征输入聚类器，并与聚类特征样本进行比较，以识别当前的试剂状况，并将其用作指导生产过程中异常试剂状况的反馈信息。通过实验分析和比较了监督信息量和特征质量对聚类结果的影响。