基于深度学习的运煤图像处理应用研究  

Research on application of coal transportation image processing based on deep learning

作　　者：陈意 张小亮 李静鹏 CHEN Yi;ZHANG Xiaoliang;LI Jingpeng(Guoneng Fengcheng Power Generation Co.,Ltd.,Fengcheng 331100,China)

机构地区：[1]国能丰城发电有限公司,江西丰城331100

出　　处：《电气应用》2024年第4期29-35,共7页Electrotechnical Application

摘　　要：人工智能技术正在高速发展,在各个领域的实际应用也变得越来越普遍。为保障发电厂运行设备的安全性,提出了基于深度学习的煤炭运煤图像检测算法,改进了基于Retinex的目标检测方法。针对图像光照不均的特点,将获取的低光图像作为网络的训练数据,采用深度曲线估计此参数矩阵,该矩阵与原始图像相乘,经过多次迭代后得到增强的图像。最后应用于公司燃运系统,检测未佩戴安全帽、未穿长裤和煤仓堵煤溢煤现象,验证了所提改进算法的正确性和实用性。Artificial intelligence technology is developing rapidly and its practical applications in various fields are becoming increasingly common.To ensure the safety of operating equipment in power plants,this paper proposes a deep learning based coal transportation image detection algorithm and improves the Retinex based object detection method.In response to the uneven lighting characteristics of the image,a deep curve is used to estimate this parameter matrix for obtaining low light images as training data for the network.This matrix is multiplied by the original image and after multiple iterations,an enhanced image is obtained.Finally,it was applied to the fuel transportation system of a company to detect the phenomena of not wearing a safety helmet,not wearing long pants and coal blockage or overflow in the coal bunker,verifying the correctness and practicality of the proposed improved algorithm.

关 键 词：人工智能 图像检测 目标检测 深度曲线 参数矩阵 

分 类 号：TM621[电气工程—电力系统及自动化] TP18[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程]