• 论文
主办单位:煤炭科学研究总院有限公司、中国煤炭学会学术期刊工作委员会
基于几何特征约束的煤矸DE-XRT精准识别方法
  • Title

    Accurate identification method of coal and gangue based on geometric feature constraints by DE-XRT

  • 作者

    何磊郭永存支亚王爽李德永胡坤程刚

  • Author

    HE Lei;GUO Yongcun;ZHI Ya;WANG Shuang;LI Deyong;HU Kun;CHENG Gang

  • 单位

    安徽理工大学 机电工程学院深部煤矿采动响应与灾害防控国家重点实验室安徽中科光电色选机械有限公司

  • Organization
    School of Mechatronics Engineering, Anhui University of Science and Technology
    State Key Laboratory of Mining Response and Disaster Prevention and Control in Deep Coal Mines
    Anhui Zhongke Photoelectric Color Sorter Machinery Co., Ltd.
  • 摘要

    双能X射线透射识别煤矸仍存在厚度、硬化、余辉和扇形效应等缺陷,面向5~150 mm宽厚度煤矸分选参数波动大、识别率低。为此,提出一种基于几何特征约束的煤矸双能X射线透射多维度识别方法。该方法通过目标图像最小外接圆直径和区域面积两个几何特征区分煤矸厚度,约束X射线透射响应特征的空间分布,进而从多个维度特征削弱缺陷影响。以少量低密度煤和高密度矸石,获取X射线透射响应特征、位置特征和几何特征,结合Relief-F特征选择建立强特征组合。检验多种分类器的识别性能,选取中等高斯SVM作为多维度方法的分类模型。以强特征组合作为输入,自动创建最终决策模型并分类未知煤矸像素点,通过像素变换图像处理方法获取分选参数 p 值。结果显示, p 值与煤矸密度呈强线性相关,利用密度可选取 p 值调控分选。而 p 值与煤矸厚度呈现弱线性相关,宽厚度范围内 p 值离散程度小、可分性好,赋予分选参数较大调整空间。批量试验验证结果显示,多维度法预排矸分选参数 p 值为33.01%,以此分选参数对不同密度、不同煤种煤矸识别,整体识别率达99.57%。对5~150 mm厚度范围原煤预排矸整体识别率达99.37%。相比较H-L法、R-L法,多维度法识别率更高,面向不同厚度煤矸计算得到的 p 值精度高、一致性更好。印证了几何特征约束下多维度识别方法的有效性及分选参数调控优势,为现有双能X射线煤矸分选装置识别算法提供了设计参考。

  • Abstract

    The dual energy X-ray transmission identification of coal gangue still faces challenges in thickness, hardening, afterglow, and fan-shaped effects, among which the parameters for 5-150 mm wide thickness coal gangue separation fluctuate significantly and the recognition rate is to be improved. Therefore, this paper proposes a multi-dimensional identification method of dual-energy X-ray transmission of coal gangue based on geometric feature constraints. This method distinguishes the thickness of coal gangue by two geometric features of the minimum circumscribed circle diameter and area of the target image, restricts the spatial distribution of X-ray transmission response characteristics, and then weakens the influence of defects from multiple dimensions. With a small amount of low-density coal and high-density gangue, the paper obtains X-ray transmission response characteristics, position characteristics, and geometric characteristics, and combine them with Relief-F feature selection to establish a strong feature combination. To test the recognition performance of multiple classifiers, medium Gaussian SVM is selected as the classification model for multi-dimensional methods. Taking strong feature combinations as input, the final decision model and classification of unknown coal gangue pixels are automatically created, and the separation parameter p-value is obtained through pixel transformation image processing method. The results show that there is a strong linear correlation between p-value and coal gangue density, and density can be used to select p-value to regulate sorting. The p-value shows a weak linear correlation with the thickness of coal gangue. Within a wide thickness range, the p-value has a small degree of dispersion and good separability, giving separation parameters a large adjustment space. The mass experimental verification results show that the p-value of the multi-dimensional method for pre discharge gangue separation parameter is 33.01%. Using this separation parameter to identify coal gangue with different densities and coal types, the overall recognition rate reaches 99.57%. The overall recognition rate of raw coal pre discharge gangue in the thickness range of 5-150 mm is 99.37%. Compared with H-L method and R-L method, the multi-dimensional method has higher recognition rate, and the p-value calculated for coal gangue with different thickness has higher accuracy and better consistency. It demonstrates the effectiveness of multi-dimensional recognition methods under geometric feature constraints and the advantages of separation parameter regulation, which is of design reference meaning for current dual energy X-ray coal gangue separation device recognition algorithms.

  • 关键词

    煤矸识别双能 X 射线几何特征多维度分选参数

  • KeyWords

    coal gangue recognition;dual-energy X-ray;geometric feature;multi-dimension;sorting parameter

  • 基金项目(Foundation)
    安徽理工大学博士研究生创新基金资助项目(2022CX1006);安徽省高校优秀青年科研资助项目(2022AH020056);国家自然科学基金面上资助项目(52274152)
  • DOI
  • 引用格式
    何 磊,郭永存,支 亚,等. 基于几何特征约束的煤矸DE-XRT精准识别方法[J]. 煤炭科学技术,2024,52(5):262−275.
  • Citation
    HE Lei,GUO Yongcun,ZHI Ya,et al. Accurate identification method of coal and gangue based on geometric feature constraints by DE-XRT[J]. Coal Science and Technology,2024,52(5):262−275.
  • 相关文章
  • 图表

    Table1

    训练集部分样本特征提取结果
    训练样本 标签 R Gl Gh αl αh Tl Th Y/px S/px D/px
    1 1.34 125.51 107.92 0.59 0.44 86.04 69.16 28.05 522 42
    1 1.36 83.35 62.00 1.18 0.87 131.76 110.89 198.89 1968 76
    1 1.27 47.08 32.90 1.90 1.50 160.35 146.74 284.67 6530 124
    1 1.22 26.98 18.27 2.72 2.23 174.38 165.61 51.99 15816 196
    矸石 2 1.49 57.47 33.06 1.91 1.31 160.84 136.73 279.89 1393 56
    2 1.45 46.87 26.06 2.22 1.56 166.61 146.51 114.76 3640 114
    2 1.32 12.61 6.48 4.15 3.47 187.22 180.19 172.75 5689 120
    2 1.36 7.73 6.67 4.07 4.09 189.26 184.42 275.9 18329 318

    Table2

    对照组分类器选择及测试结果
    方法 分类器 训练集样品数 煤样品数 测试结果/%
    H-L 二次SVM 267×3 126×3 98.1
    R-L 中等KNN 267×3 126×3 98.1
    多维度 中等高斯SVM 267×11 126×11 98.5

    Table3

    煤矸像素点对应多维特征
    像素点特征参数 R Gl Gh αl αh Tl Th Y/px S/px D/px
    1.29 134 119 0.47 0.36 72.61 59.42 237 1508 68
    1.48 138 116 0.50 0.33 75.61 55.42 237 1508 68
    1.34 135 118 0.48 0.35 73.61 58.42 237 1508 68
    1.21 136 125 0.42 0.35 66.61 57.42 236 1508 68
    矸石 1.62 75 41 1.55 0.95 153.97 120.22 382 11958 158
    1.68 74 38 1.63 0.97 156.97 121.22 382 11958 158
    1.61 72 39 1.60 0.99 155.97 123.22 382 11958 158
    1.40 70 46 1.44 1.02 148.97 125.22 381 11958 158

    Table4

    不同方法计算 p值与密度拟合结果
    方法 煤种 拟合公式 相关系数R 密度/(g·cm−3 p p值均值
    H-L 肥煤 y=1.0132x−1.3063 0.95 1.8 51.75% 47.99%
    焦煤 y=1.1092x−1.5125 0.95 1.8 48.41%
    气煤 y=0.95815x−1.2865 0.87 1.8 43.82%
    R-L 肥煤 y=0.65233x−0.6788 0.98 1.8 49.54% 49.61%
    焦煤 y=0.78235x−0.92803 0.99 1.8 48.02%
    气煤 y=0.99932x−1.286 0.95 1.8 51.28%
    多维度 肥煤 y=0.70215x−0.90147 0.98 1.8 36.34% 33.01%
    焦煤 y=0.75278x−1.0224 0.96 1.8 33.26%
    气煤 y=0.61043x−0.80354 0.89 1.8 29.52%

    Table5

    不同方法计算 p值与厚度拟合关系
    方法 类别 拟合公式 相关系数
    H-L y=0.0059x− 0.164 0.82
    矸石 y=0.0004x+ 0.939 0.92
    R-L y=0.0006x+ 0.011 0.94
    矸石 y=−0.0006x+ 0.965 0.48
    多维度 y=0.0004x+ 0.011 0.59
    矸石 y=0.0001x+ 0.951 0.56

    Table6

    H-L法批量验证结果
    煤样 肥煤 焦煤 气煤 整体情况
    目标个数  113 182 171 466
    煤的个数 58 88 81 227
    矸石个数 55 94 90 239
    煤错分 0 0 0 0
    矸石错分 0 2 6 8
    总错分 0 2 6 8
    总体识别率/% 100.00 98.90 96.49 98.28
    煤识别率/% 100.00 100.00 100.00 100.00
    矸石识别率/% 100.00 97.87 93.33 96.65

    Table7

    R-L法批量验证结果
    煤样 肥煤 焦煤 气煤 整体情况
    目标个数  113 182 171 466
    煤的个数 58 88 81 227
    矸石个数 55 94 90 239
    煤错分 0 0 0 0
    矸石错分 5 3 9 17
    总错分 5 3 9 17
    总体识别率/% 95.58 98.35 94.74 96.35
    煤识别率/% 100.00 100.00 100.00 100.00
    矸石识别率/% 90.91 96.81 90.00 92.89

    Table8

    多维度法批量验证结果
    煤样 肥煤 焦煤 气煤 整体情况
    目标个数  113 182 171 466
    煤的个数 58 88 81 227
    矸石个数 55 94 90 239
    煤错分 0 1 1 2
    矸石错分 0 0 0 0
    总错分 0 1 1 2
    总体识别率/% 100.00 99.45 99.42 99.57
    煤识别率/% 100.00 98.86 98.77 99.12
    矸石识别率/% 100.00 100.00 100.00 100.00

    Table9

    批量宽厚度煤矸识别验证
    方法 样品 目标
    个数
    煤错
    分数
    矸石
    错分数
    总错
    分数
    煤识
    别率/%
    矸石识
    别率/%
    总体识
    别率/%
    H-L 272 12 0 12 95.59 100 97.47
    矸石 202
    R-L 272 4 0 4 98.53 100 99.16
    矸石 202
    多维度 272 3 0 3 98.90 100 99.37
    矸石 202
相关问题

主办单位:煤炭科学研究总院有限公司 中国煤炭学会学术期刊工作委员会

©版权所有2015 煤炭科学研究总院有限公司 地址:北京市朝阳区和平里青年沟东路煤炭大厦 邮编:100013
京ICP备05086979号-16技术支持: 云智互联
Baidu
map