基于改进GMM背景差分方法的蒸汽泄漏检测研究
投稿时间:2022-10-24  修订日期:2022-11-22  点此下载全文
引用本文:
摘要点击次数: 147
全文下载次数: 0
作者单位邮编
鄢家鑫* 中国核动力研究设计院 610213
中文摘要:为检测反应堆回路试验装置蒸汽泄漏,本文提出了一种改进的基于背景差分法的蒸汽泄漏红外视频检测方法。该方法利用红外相机采集图像,使用Wasserstein距离匹配像素点,采用K-means方法去更新改进的混合高斯背景模型(GMM);在后处理中,采用自适应均值滤波及形态学方法抑制环境噪声,通过对泄漏图像灰度形状面积的分层融合特征判断实现对蒸汽泄漏的分级检测。为实现对算法效果的量化,利用泄漏报警率、F1分数等指标评价检测方法的优劣,提出以Dice系数作为算法分割效果的评价函数;在steamTS200inf数据集上的试验结果表明,泄漏报警率可达98.52%,最后一帧分割Dice值可达78.31%,本文方法可有效检测蒸汽泄漏。
中文关键词:泄漏检测  背景差分  红外视频  图像分割
 
Research on Steam Leak Detection Based on Improved GMM Background Difference Method
Abstract:In order to detect the steam leakage of the reactor loop test device, the paper proposed an improved infrared video detection method of steam leakage based on the background difference method. This method collects images with the infrared camera, and uses the Wasserstein distance to match the pixels, and uses the K-means method to update the improved Gaussian mixture background model (GMM). In the post-processing, adaptive mean filtering and morphological methods are used to suppress environmental noise. The hierarchical fusion feature judgment of shape and area realizes the hierarchical detection of steam leakage. In order to quantify the effect of the algorithm, the leakage alarm rate, F1 score and other indicators are used to evaluate the pros and cons of the detection method, and the Dice coefficient is proposed as the evaluation function of the segmentation effect of the algorithm; the experimental results on the steamTS200inf dataset show that the leakage alarm rate can reach 98.52%, and the Dice value of the last frame can reach 78.31%. The method in this paper can effectively detect steam leakage.
keywords:YAN Jiaxin,CHEN Qing, LI Pengzhou, ZHOU Han, LIU Xiaodong
查看全文   查看/发表评论   下载pdf阅读器