环境适应能力强、识别精度高是有效分离煤和矸石的前提。采用双能X射线透视煤和矸石并成像，避免了粉尘、光照强度和物料表面等外界因素影响。但双能X射线探测器采集射线能量数据存在余晖效应、厚度效应和射束硬化效应等缺陷。为降低缺陷影响，提高煤和矸石识别率，提出一种联合R值图像与高、低能图像特征对煤和矸石进行多维度分析的方法。首先基于双能X射线采集系统获取高、低能图像，并利用比值法得到R值图像；然后针对所获取的三种图像，研究煤和矸石密度及灰分含量等关键物性参数与图像特征关系，据此设计特征提取方案，共计提取8个特征参量，形成一种强特征组合；最后采用Relief算法度量每个特征参量的重要性，进而设计分类试验。以不同地区肥煤、焦煤、气煤和矸石为试验对象，观察剔除权重较低的特征后，分类模型准确率，发现以特征组合[ Rc, μlc, μl, R]为输入，PSO-SVM分类模型对三种煤混合矸石识别效果最佳，识别率为99.4%。结合PSO-SVM分类模型和[ Rc, μlc, μl, R]的特征组合对肥煤、焦煤和气煤分别混合矸石进行识别验证，结果表明：肥煤混合矸石识别率为98.89%，焦煤混合矸石识别率为100%，气煤混合矸石的识别率为99.44%。本方法通过联合多张图像的多个特征，进行多维度分析，发现R值图像特征和高能图像特征对煤和矸石的区分度最好，可有效降低双能X射线缺陷影响，能以较少的特征，实现对不同煤种的较高识别率，优于现有方法。此外，以灰分、密度为参照选取特征阈值，贴合实际需求，减少了工程应用中根据矿区煤质差异进行参数调整的频次，提高了识别模型的泛化能力。

There are the prerequisites for effective separation of coal and gangue on strong environmental adaptability and high identification accuracy. Dual-energy X-ray is used to see through coal and gangue and image, which avoids the influence of external factors such as dust, light intensity and material surface. However, the radiation energy data collected by dual-energy X-ray detectors have defects such as afterglow effect, thickness effect and beam hardening effect. In order to reduce the influence of defects and improve the recognition rate of coal and gangue, a multi-dimensional analysis method of coal and gangue is proposed by combining R-value images with high and low energy image features. Firstly, high-energy and low-energy images are obtained based on dual-energy X-ray acquisition system, and R-value images are obtained by ratio method. Then, according to the three images obtained, the relationship between key physical parameters such as coal and gangue density and ash content and image features is studied, and a feature extraction scheme is designed accordingly. A total of 8 feature parameters are extracted to form a strong feature combination. Finally, relief algorithm is used to measure the importance of each feature parameter, and then classification experiments are designed. Taking rich coal, charred coal, gas coal and gangue from different areas as experimental objects, the accuracy of classification model was observed after removing the features with lower weight. It was found that PSO-SVM classification model had the best recognition effect on the mixed gangue of three kinds of coal with a recognition rate of 99.4% by taking the feature combination [Rc, μlc, μl, R] as input. Combined with the PSO-SVM classification model and the characteristic combination of [Rc, μlc, μl, R], the identification and verification of mixed gangue of rich coal, charred coal and gas coal are carried out respectively. The results show that the identification rate of mixed gangue of rich coal is 98.89%, that of mixed gangue of charred coal is 100%, and that of mixed gangue of gas coal is 99.44%. Combined multiple features of multiple images for multi-dimensional analysis, the method finds that R-value image features and high-energy image features have the best discrimination degree to coal and gangue. What’s more, it can effectively reduce the influence of dual-energy X-ray defects, and realize higher recognition rate to different coal types with fewer features, which is superior to the existing methods. In addition, the ash content and density are taken as the reference to select the characteristic threshold, which meets the actual demand, reduces the frequency of parameter adjustment according to the coal quality difference in the mining area in engineering application, and improves the generalization ability of the identification model.

1 图像获取与缺陷分析

　　1.1图像获取

　　1.2图像缺陷分析 

2 特征提取方法 

　　2.1 R值均值 

　　2.2基于R值的矸石含量特征

　　2.3高低能图像灰度、纹理特征 

　　2.4基于高、低能图像灰度的矸石含量特征

3 Relief特征选择和分类模型

4 讨论与结果分析

　　4.1模型分类结果分析

　　4.2 不同方法对比分析

5 试验验证 

6 结论