摘要
为解决现有露天矿电铲铲齿缺失状态靠人员观察进行识别的现状,构建一套电铲铲齿缺失状态的实时智能监测系统。该系统基于Caffe的深度学习框架,选择Faster RCNN算法,将铲齿模拟图片分为缺陷和完整2类,并分别对2类图像进行加噪、放缩、旋转、亮度调节等操作,以模拟现场复杂环境对采集样本的影响;为缩短样本图片检测时间,更好地实现系统的实时监测,对Faster RCNN算法中Proposal检测阶段进行改进,构建了电铲铲齿智能监测系统;通过对卷积神经网络和训练模型不断调整和修改,使得缺陷铲齿和正常铲齿的正确识别率达到86.68%,网络损失率也趋于稳定。研究方法可为深度学习应用于矿山的智能视频监控、安全行为和安全状态的智能识别研究提供借鉴。
In order to solve the status quo of identifying the missing state of the shovel teeth by personnel observation in the existing opencast,a real-time intelligent monitoring system for the missing state of shovel teeth was constructed.The system was based on the deep learning framework of Caffe and selects the Faster RCNN algorithm,which divided the simulation pictures of shovel teeth into two kinds of defects and complete types.In order to simulate the effect of the complex environment on the sample collection,the two types of images were processed by adding noise,mitigation,amplification,rotating and brightness adjustment respectively.In order to shorten the sample image detection time and achieve real-time monitoring of the system better,the Proposal detection stage in the Faster RCNN algorithm was improved,and an electric shovel shovel tooth intelligent monitoring system was constructed.Through the continuous adjustment and modification,the correct recognition rate of the defective shovel teeth and the normal shovel teeth reached 86.68%,and the network loss rate tends to be stable.The research method also provides reference for deep learning to be applied to intelligent video surveillance of mines,safe behaviors and intelligent identification of safety state.
作者
周世禄
杨小彬
王逍遥
李娜
ZHOU Shilu;YANG Xiaobin;WANG Xiaoyao;LI Na(School of Resource and Safety Engineering,China University of Mining and Technology(Beijing),Beijing 100083,China)
出处
《煤炭科学技术》
CAS
CSCD
北大核心
2020年第S01期119-124,共6页
Coal Science and Technology
基金
国家自然科学基金资助项目(50904071,51274207).
关键词
深度学习
智能监测
卷积神经网络
Caffe框架
电铲铲齿
deep learning
intelligent monitoring
convolutional neural network
Caffe framework
shovel teeth
作者简介
周世禄(1992—),男,河北沧州人,硕士研究生。E-mail:zhoushilu00@126.com;通讯作者:杨小彬(1976—),男,重庆人,副教授,博士生导师,博士。Tel:yangxiaobin02@126.com
