RetinaNet图像识别技术在煤矿目标监测领域的应用研究  

Application research of RetinaNet image recognition technology in coal mine target monitoring

作　　者：谭章禄[1] 陈孝慈 TAN Zhanglu;CHEN Xiaoci(School of Management,China University of Mining and Technology(Beijing),Beijing 100083,China)

机构地区：[1]中国矿业大学(北京)管理学院,北京100083

出　　处：《矿业安全与环保》2020年第5期65-70,76,共7页Mining Safety & Environmental Protection

基　　金：国家自然科学基金项目(61471362)。

摘　　要：为了解决当前煤矿监控中存在的人工干预多、监测效率低等问题,建立基于RetinaNet的单阶段煤矿目标检测器,通过实验确定检测关键参数并验证检测效果。实验结果表明:RetinaNet目标检测器能够自动检测及提取人员等关键对象,整体性能可以满足煤矿监控的需求;RetinaNet目标检测器能够在较差的环境条件下实现对目标对象的准确检测,对于人员的辨识已经达到较为理想的水平;基于现有数据构建的图像识别模型,尚不能较好地识别各类煤矿机械设备。RetinaNet目标检测器相关功能的实现,有赖于建立专业图像数据集,并准确地训练模型进而发掘数据的深度价值。In order to solve the problems existing in current coal mine monitoring,such as excessive manual intervention and low monitoring efficiency,a single-stage coal mine target detector was established based on RetinaNet,and the key parameters of detection were determined through experiments and the detection effect was verified.The experimental results show that the RetinaNet target detector can automatically detect and extract key objects such as personnel,and the overall performance can meet the requirements of coal mine monitoring;RetinaNet target detector can realize accurate detection of target objects under poor environmental conditions,which has reached a relatively ideal level for human identification;the image recognition model based on the existing data can not identify various types of coal mine machinery and equipment well.The realization of relevant functions of RetinaNet target detector depends on the establishment of professional image data set and the accurate training model to explore the depth value of data.

关 键 词：矿井监控 RetinaNet 目标检测器 对象检测 识别模型 专业数据集 

分 类 号：TD76[矿业工程—矿井通风与安全] X923[环境科学与工程—安全科学]