基于对抗生成网络的矿井烟火安全预警检测算法研究  

Research on Mine Fireworks Safety Warning and Detection Algorithm Based on Adversarial Generative Network

作　　者：王哲 陈峤鹰 李青 傅哲 张斌[3] 文宇阳 WANG Zhe;CHEN Qiaoying;LI Qing;FU Zhe;ZHANG Bin;WEN Yuyang(CCTEG Shenyang Research Institute,Fushun 113122,China;CCTEG Fushun Testing Center,Fushun 113122,China;Xi’an Jiaotong University,Xi’an 710049,China;Xi’an Bossun Security Technologies Co.,Ltd.,Xi’an 710304,China;CCTEG Shanghai Test Technology Co.,Ltd.,Shanghai 201401,China)

机构地区：[1]中煤科工集团沈阳研究院有限公司,辽宁抚顺113122 [2]抚顺中煤科工检测中心有限公司,辽宁抚顺113122 [3]西安交通大学,陕西西安710049 [4]西安博深安全科技股份有限公司,陕西西安710304 [5]上海煤科检测技术有限公司,上海201401

出　　处：《煤矿机电》2024年第6期7-11,28,共6页Colliery Mechanical & Electrical Technology

摘　　要：近年来,煤矿安全一直是重点关注的领域,各种技术不断为煤矿生产安全提供更好的保障。传统视频监控系统依赖人力,存在视野盲区和效率低下的问题。为了提高煤矿安全监控的智能化水平,提出了一种基于海思Hi3559A边缘计算设备和深度学习技术的解决方案。该方案通过生成对抗网络(GAN)生成训练样本,并对YOLOv7检测算法进行优化,引入滑动窗口自注意力机制(Swin Transformer)、双路特征提取网络(BiBranchBackbone)和网格特征融合模块(GrideFuse),提升了检测和分割精度,以实现复杂场景下明火和烟雾的精确检测和分割。试验结果表明,改进后的YOLOv7算法在检测框mAP和分割mAP上分别提高了2.7%和2.3%。提出的智能煤矿安全预警监测系统已在实际应用中取得了显著效果,为煤矿企业的安全生产提供了有力保障。In recent years,coal mine safety has always been a key area of concern,and various technologies continuously provide better guarantees for coal mine production safety.Traditional video surveillance systems rely on manpower and suffer from blind spots and low efficiency.In order to improve the intelligent level of coal mine safety monitoring,a solution based on Hi3559A edge computing equipment and deep learning technology was proposed.This scheme generates training samples through Generative Adversarial Networks(GANs) and optimizes the YOLOv7 detection algorithm.A sliding window self attention mechanism(SwinTransformer),a dual feature extraction network(BiBranchBackbone),and a grid feature fusion module(GrideFuse) was introduced to improve detection and segmentation accuracy.To achieve precise detection and segmentation of open flames and smoke in complex scenes.The experimental results showed that the improved YOLOv7 algorithm had improved detection box mAP and segmentation mAP by 2.7% and 2.3%,respectively.The proposed intelligent coal mine safety warning and monitoring system has achieved significant results in practical applications,providing strong guarantees for the safety production of coal mining enterprises.

关 键 词：智能系统 YOLOv7 边缘计算设备 实时检测 智能化分析 

分 类 号：TD679[矿业工程—矿山机电]