摘要
在煤炭开采过程中,对煤矸进行有效识别不仅可以提高煤炭的燃烧质量,还可以降低污染物排放,现有传统的带式输送机煤矸识别技术操作复杂、效率低、成本高。针对上述问题,提出一种基于YOLOv8-ER的带式输送机煤矸目标检测模型。该模型在YOLOv8基础模型上,引入高效通道注意力网络,对YOLOv8的主干网络进行优化,以增强图像中的重要信息、抑制次要信息,加强主干网络的特征提取能力,降低模型计算复杂度,提升模型学习效率和检测精度。该模型还引入结构重参化技术,以简化推理过程中的网络架构,进一步降低网络体积,增强模型在深层网络中的训练稳定性,提升模型推理速度。实验结果表明,YOLOv8-ER在处理速度和检测精度上均有显著提升,实现了240.8帧/s的高速检测,m AP@0.5达到了92.3%的高准确率;与传统YOLOv8模型相比,准确率提高了3.2%,m AP@0.5提高了4.4%,PR平衡率提高了2.6%,浮点运算次数和GPU使用量分别降低了23.2%和18.6%。根据煤矸可视化结果可以得出,YOLOv8-ER模型能更精准地识别煤矸,避免误检漏检现象,且置信度更高。
[Objective]The primary objective of this study is to tackle the manifold challenges associated with identifying coal gangue during coal mining,including but not limited to the inefficiencies,high operational complexity,and substantial costs inherent in traditional identification methods.By integrating and refining advanced deep learning technologies,particularly the YOLOv8-based object detection model,we seek to develop an intelligent system capable of automatically,rapidly,and precisely identifying coal gangue.This intelligent system will not only enhance the efficiency and quality of coal mining and minimize human errors and cost inefficiencies but also foster the sustainable utilization of coal resources,mitigate environmental pollution,and provide technical support for the green transition of the coal industry.[Methods]To achieve the aforementioned research objectives,this study adopted a systematic research method.First,we conducted an in-depth analysis of the advantages and disadvantages of the YOLOv8 model and proposed the YOLOv8-ER model based on this.This model deeply optimized the backbone network of YOLOv8 by introducing an efficient channel attention network,enabling the model to focus more effectively on key features in the image and improve the accuracy and efficiency of feature extraction.At the same time,we applied structural reparameterization technology to simplify the network architecture of the model,reduce the computational complexity and resource consumption,and improve the inference speed and stability of the model.In the model training phase,we used a large-scale and diverse dataset of coal gangue images and continuously optimized the model parameters through multiple iterations of training to ensure the accuracy and robustness of the model in practical applications.[Results]This study has achieved multiple innovative results in the field of coal gangue identification.First,the YOLOv8-ER model has achieved significant improvements in both processing speed and detection accuracy.The high-speed detection capability of the YOLOv8-ER model(i.e.,240.8 frames/second)can meet the high real-time requirements of application scenarios,with a performance rate of up to 92.3%mAP@0.5,which ensures the accuracy and reliability of recognition.Second,compared with the traditional YOLOv8 model,YOLOv8-ER has made breakthroughs in multiple key indicators,such as an accuracy improvement of 3.2%,mAP@0.5 improved by 4.4%,and PR balance rate improved by 2.6%.These improvements make the model more competitive in practical applications.In addition,by optimizing the network structure and applying reparameterization techniques,we have successfully reduced the computational complexity and resource consumption of the model,improving its operational efficiency and economy.[Conclusions]In summary,this study successfully developed an efficient and accurate coal gangue identification technology by introducing and optimizing the YOLOv8-ER model.This technology not only overcomes the limitations of traditional recognition methods and improves the efficiency and quality of coal mining but also provides strong support for the intelligent transformation of the coal industry.In the future,with the continuous advancement of technology and the expansion of application scenarios,the YOLOv8-ER model is expected to play an important role in more fields,promoting the sustainable development and innovation of related industries.At the same time,we will continue to conduct in-depth research on the optimization and extended application of this model to meet the needs of coal gangue recognition in different scenarios.
作者
鲁杰
王劭琛
魏征
LU Jie;WANG Shaochen;WEI Zheng(College of Coal Engineering,Shanxi Datong University,Datong 037003,China;College of Mining Engineering,China University of Mining and Technology,Xuzhou 221116,China)
出处
《实验技术与管理》
CAS
北大核心
2024年第10期67-73,共7页
Experimental Technology and Management
基金
山西省基础研究计划项目(202303021222203)
山西大同大学2023年科研创新项目(23CX54)。