融合自注意机制的矿工行为识别算法  

Miner Behavior Recognition Algorithm with Self-attention Mechanisms

在线阅读下载全文

作  者:赵月爱[1,2] 郝慧琦 王玲 ZHAO Yue-ai;HAO Hui-qi;WANG Ling(School of Computer Science and Technology,Taiyuan Normal University,Jinzhong 030602,China;Shanxi Key Laboratory of Intelligent Optimization Computing and Blockchain Technology,Jinzhong 030619,China;School of Automation and Software,Shanxi University,Taiyuan 030006,China)

机构地区:[1]太原师范学院计算机科学与技术学院,山西晋中030602 [2]山西智能优化计算与区块链技术重点实验室,山西晋中030619 [3]山西大学自动化与软件学院,山西太原030006

出  处:《计算机技术与发展》2025年第4期186-192,共7页Computer Technology and Development

基  金:山西省自然科学研究面上项目(202303021221173)。

摘  要:我国煤矿工业建设已逐步转化为智能化模式,采用深度学习算法识别煤矿监控视频中矿工不安全行为具有十分重要的意义。但煤矿井下环境恶劣,提取矿工行为特征干扰大,目前目标检测算法存在精度低、计算量大的问题。针对这些挑战,该文提出了一种融合自注意机制的矿工行为识别模型CMFA-YOLO,采用ConvNeXt V2模块来替代原有YOLOv8n中的C2f模块,实现了Transformer思想与目标检测领域的完美融合,有效降低了模型计算量。进一步采用融合边界框损失函数MF-IoU(MPD-and-Focaler IoU)提高模型的精度。此外引入了自适应空间相关性金字塔注意力ASCPA,配合ConvNeXt V2发挥模型最佳效果。经过多项实验结果显示,CMFA-YOLO在煤矿视频监控场景下的检测精度达到99.0%,为煤矿场景下的异常行为检测任务提供高精确算法。The construction of China's coal mining industry has gradually moved to an intelligent mode,and the use of deep learning algorithms to detect the unsafe behavior of miners in coal mine surveillance videos is of great significance.However,the underground environment of coal mines is harsh,the extraction of miners'behavioral features is highly disturbed,and the current target detection algorithms have the problems of low accuracy and large computational volume.In response to these challenges,we propose a miner behavior recognition model CMFA-YOLO that incorporates a self-attention mechanism.By using the ConvNeXt V2 module to replace the C2f module in the original YOLOv8n,the model achieves a perfect integration of Transformer with the field of object detection,effectively reducing the computational load of the model.The fusion bounding box loss function MF-IoU(MPD-and-Focaler IoU)is further adopted to improve the accuracy of the model.In addition,the adaptive spatial correlation pyramid attention ASCPA is introduced to work with ConvNeXt V2 to give the best effect.Experimental results show that CMFA-YOLO achieves 99.0%detection accuracy in coal mine video surveillance scenarios,providing highly accurate algorithms for the task of detecting abnormal behavior in coal mine scenarios.

关 键 词:煤矿监控检测 不安全行为识别 目标检测 自注意力机制 融合边界框损失函数 

分 类 号:TP39[自动化与计算机技术—计算机应用技术] TD76[自动化与计算机技术—计算机科学与技术] TD67[矿业工程—矿井通风与安全]

 

参考文献:

正在载入数据...

 

二级参考文献:

正在载入数据...

 

耦合文献:

正在载入数据...

 

引证文献:

正在载入数据...

 

二级引证文献:

正在载入数据...

 

同被引文献:

正在载入数据...

 

相关期刊文献:

正在载入数据...

相关的主题
相关的作者对象
相关的机构对象