郭龙真, 朱朋飞,等.基于机器视觉的综采工作面支架推、移状态识别[J].河南理工大学学报（自然科学版）,2026,45(5):116-124.

GUO L Z, ZHU P F,et al.Recognition of hydraulic support push-pull state in fully mechanized longwall mining face based on machine vision[J].Journal of Henan Polytechnic University(Natural Science) ,2026,45(5):116-124.

基于机器视觉的综采工作面支架推、移状态识别

郭龙真1,2, 朱朋飞3

1.河南理工大学 电气工程与自动化学院，河南 焦作  454003；2.郑州煤矿机械集团股份有限公司 博士后工作站，河南 郑州 450000；3.郑州煤机智控技术创新中心有限公司，河南 郑州  450000

摘要: 目的针对煤矿井下综采工作面支架推溜和拉架状态主要依赖人工巡检和传感器监测、存在效率低和可靠性差的问题，本文提出一种基于机器视觉的支架状态识别方法，以实现工作面全实时感知和智能化运维。  方法该方法采用自顶向下的方式实现对支架底座的关键点检测。目标检测阶段利用YOLOv8m网络进行特征提取和处理，得到支架底座在图像坐标系中的位置；依据当前检测目标信息确定支架底座在图像坐标系中的空间分布，生成相应的空间置信度增强矩阵，将空间置信度矩阵附加到YOLOv8m网络Head输出层进行相应后处理，得到图像中准确的目标信息。关键点检测阶段通过设计支架底座关键点数据集，并对Lite-HRNet网络输出结构进行调整，训练得到支架底座的关键点识别模型，实现上一阶段支架底座的关键点提取。最后，统计与分析视频序列中同一支架底座中关键点位置变化，确定当前支架的运动状态。  结果结果表明：本文方法能够高效、准确识别工作面推溜、拉架动作状态，且识别准确率高达88.29%。  结论该方法实用价值较高，可为井下煤矿生产提供强有力的技术支持，助力煤矿行业实现智能化转型与升级。

关键词:智能识别;空间置信度矩阵;关键点检测;推溜;拉架

doi:10.16186/j.cnki.1673-9787.2024120047

基金项目:国家自然科学基金资助项目（61972016）；中国国家铁路集团有限公司科研项目（N2023X005）

收稿日期:2024/12/19

修回日期:2025/09/08

出版日期:2026-07-24

Recognition of hydraulic support push-pull state in fully mechanized longwall mining face based on machine vision

Guo Longzhen1,2, Zhu Pengfei3

1.School of Electrical Engineering and Automation， Henan Polytechnic University， Jiaozuo  454003， Henan， China；2.Zhengzhou Coal Mining Machinery Group Co.， Ltd.， Postdoctoral Research Station， Zhengzhou  450000， Henan， China；3.Zhengzhou ZMJ Intelligent Control Technology Innovation Center Co.， Ltd.， Zhengzhou  450000， Henan， China

Abstract: Objectives Addressing the issues of low efficiency and poor reliability caused by the heavy reliance on manual inspection and sensor monitoring for the push-and-pull states of hydraulic supports in fully mechanized coal mining faces， this paper proposes a support state recognition method based on machine vision to achieve real-time perception and intelligent operation and maintenance of the working face. Methods The proposed method adopts a top-down framework for keypoint-based analysis of hydraulic support bases. In the target detection stage， the YOLOv8m network is employed to extract visual features and obtain the positions of support bases in the image coordinate system. Based on the detected results， the spatial distribution of supports is analyzed， and a Spatial Confidence Enhancement Matrix （SCEM） is constructed to improve localization reliability. The SCEM is integrated into the YOLOv8m detection head for refined post-processing to enhance detection accuracy. In the keypoint detection stage， a dedicated dataset of hydraulic support base keypoints is constructed， and the network architecture of Lite-HRNet is modified to train a keypoint detection model. Finally， the motion state of each support is determined by analyzing temporal variations of keypoint positions in video sequences.  Results Experimental results demonstrate that the proposed method can accurately recognize the push-pull states of hydraulic supports in fully mechanized mining faces， achieving a recognition accuracy of 88.29%. Conclusions The proposed method shows strong practical applicability and provides effective technical support for underground coal mining operations， promoting the intelligent transformation and upgrading of the coal mining industry.

Key words:intelligent recognition;spatial confidence enhancement matrix;keypoint detection;hydraulic support;push-pull state