高斯混合模型在VOCs泄漏自动识别中的研究

LDAR(leak detection and repair)技术是石化企业减排降本的重要手段,通过定期检测泄漏点,及时发现、修复、替换泄漏组件,从而减少VOCs(volatile organic compounds)的泄漏排放。然而在使用红外摄像仪进行泄漏识别时,单凭人眼的观察难以快速、准确确定泄漏点位置。为此,实时、高效的高斯混合模型被首次用来辨识化工泄漏。首先通过高斯混合模型,建立红外视频背景模型,并由背景模型提取出前景泄漏特征;然后运用μ-σ准则获得二值化图片,随后运用形态学开运算处理,进一步确定泄漏发生的位置信息;最后,采用K-means++算法对泄漏信息进行聚类框选,并把框选结果在原视频中展现,帮助现场检测人员快速确定泄漏发生的位置。实践证明,该方法对红外视频中泄漏发生的位置具有较高的辨识准确率。

VOCs Leakage Auto-identification Algorithm Based on Gaussian Mixture Model

LDAR(leak detection and repair) technology is an important method for petrochemical enterprises to reduce emissions and costs. It can detect, repair and replace leakage components in a timely manner through regular detection of leakage points, so as to reduce the leakage emissions of VOCs(volatile organic compounds). However, when using infrared camera for leak identification, it is difficult to determine the location of leakage point quickly and accurately by the observation of human eyes alone. Therefore, the real-time efficient Gaussian mixture model is first used to identify chemical production leakage. Firstly, the background model of infrared video is established by Gaussian mixture model, and the foreground leakage characteristics are extracted from the background model. Then, the two-value image is obtained by using the μ-σ criterion and the location information of leakage is further determined by morphological open operation. Finally, K-means++ algorithm is used to cluster the leakage information and displays the identification results in the original video to help field inspectors quickly determine the location of the leakage. Practice shows that this method has a high identification accuracy for the location of leakage in infrared videos.