基于视频序列的火焰识别模型研究  

Flame Recognition Model based on Video Sequence

作　　者：杨俊 YANG Jun(China Energy Dadu River Zhentouba Power Generation Co., Ltd., Leshan 614000, China)

机构地区：[1]国能大渡河枕头坝发电有限公司,四川乐山614000

出　　处：《水电与新能源》2022年第6期29-31,共3页Hydropower and New Energy

摘　　要：枕头坝电站利用高效的图像处理算法,建立火焰识别模型,并增加LSTM网络模块,对厂区已安装的工业电视摄像头采集的图像和视频提取有效图像,记录视频前后时序特征,进行视频动态目标的处理,完成对异物的识别,排除了厂区复杂环境的影响。模型指标精确率达到96%,召回率为93%,准确率94%,满足水电站复杂环境的要求。The flame recognition model developed in Zhentouba Hydropower Station is introduced.Adopting the effective image processing algorithm and the LSTM network module,valid images are extracted from the images and videos collected by the industrial TV cameras installed in the station area.The prior and post time sequence characteristics of the video are recorded and the dynamic objects are recognized.Also,irrelevant objects are identified and the influence of the complex environment of the station area is eliminated.Practical tests show that the precision index of the model is 96%,with the recall rate of 93%and the accuracy rate of 94%,which satisfy the requirement of the complex environment of hydropower stations.

关 键 词：火焰识别 PyTorch 动态目标识别 SLIC算法 

分 类 号：TV737[水利工程—水利水电工程]