基于Rs-BasicVSR的煤机装备视频超分辨率算法  

Video Super-resolution Algorithm for Coal Mining Machinery Based on Rs-BasicVSR

作　　者：徐慈强 贾运红 田原 XU Ciqiang;JIA Yunhong;TIAN Yuan(China Coal Research Institute,Beijing 100013,China;CCTEG Taiyuan Research Institute Co.,Ltd.,Taiyuan 030032,China;China National Engineering Laboratory for Coal Mining Machinery,Taiyuan 030006,China)

机构地区：[1]煤炭科学研究总院,北京100013 [2]中国煤炭科工集团太原研究院有限公司,太原030032 [3]煤矿采掘机械装备国家工程实验室,太原030006

出　　处：《煤炭技术》2024年第8期226-229,共4页Coal Technology

基　　金：国家重点研发计划(2020YFB1314003);山西省重点研发计划(2020XXX001);山西省重点研发计划(202202020101005)。

摘　　要：煤矿井下煤机装备搭载的摄像仪传输的视频信息存在信号干扰,分辨率低,细节缺失等问题,对后续的视频智能分析有着极大挑战。针对该问题,提出一种基于Rs-Basic VSR的煤机装备视频超分辨率算法。首先,在光流对齐部分改用RAFT网络,通过对图像特征建立相关空间,并根据特征信息进行搜索,提升光流计算精度;接着对网络结构进行裁剪,减少卷积层个数。实验结果表明,该算法在平均PSNR上可达31.83 dB,SSIM上达到了0.8919,平均推理时间为60 ms,模型参数量为5.9×10~6,主观视觉效果和客观指标上可以满足煤矿井下视频处理需求。The video information transmitted by the camera equipped with the coal machine in the coal mine has problems such as signal interference,low resolution,and lack of details,which poses great challenges to the subsequent intelligent analysis of video.To solve this problem,a super-resolution algorithm for coal machine equipment video based on Rs-Basic VSR is proposed.Firstly,the RAFT network is used in the optical flow alignment part,and the correlation space is established for the image features,and the search is carried out according to the feature information to improve the optical flow calculation accuracy.Then the network structure is clipped to reduce the number of convolutional layers.The experimental results show that the proposed algorithm can reach 31.83 dB on average PSNR,0.8919 on SSIM,the average inference time is 60 ms,and the number of model parameters is 5.9×106,which can meet the requirements of underground video processing in coal mines in terms of subjective visual effects and objective indicators.

关 键 词：煤机装备 视频超分辨率 光流 RAFTS 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]