基于Mask R-CNN的锅炉火焰图像核心高温区域提取

Extraction of Core High Temperature Region of Boiler Flame Image Based on Mask R-CNN

为了提高锅炉火焰核心高温区域特征提取的准确性和稳定性,结合四角切圆燃烧锅炉火焰特点,提出了一种基于Mask R-CNN的火焰核心高温区域轮廓提取方法。通过某330 MW燃煤锅炉工业火焰监控系统获取约10 000帧历史运行工况的火焰图像进行建模,并对升负荷工况的火焰图像进行在线处理验证。结果表明:基于Mask R-CNN模型的核心高温区域检测精准率达到92.35%,相比于传统的阈值分割法、边缘检测法具有更高的稳定性;Mask R-CNN模型能更好地适应变负荷时脉动火焰波动、快速变化的特点,轮廓边缘受其他区域的影响较小,有助于针对火焰核心高温区域轮廓进行火焰几何特征、分布特征的提取计算。

In order to improve the accuracy and stability of feature extraction of the core high temperature region of boiler flame,a method of extracting the contour of flame core high temperature region based on Mask R-CNN(mask region-based convolutional neural network)was proposed in combination with the flame characteristics of tangential-fired boiler.About 10000 frames of historical operational flame images from the industrial flame monitoring system of a 330 MW coal-fired boiler were used for modelling,and the flame images of uploading conditions were processed and verified online.Results show that,the accuracy of core high temperature region detection based on Mask R-CNN model reaches 92.35%,which has higher stability than traditional threshold segmentation method and edge detection method.The Mask R-CNN model can better adapt to the characteristics of the fluctuation and rapid change of the pulsating flame when the load changes,and the contour edge is less affected by other regions,which is conducive to the extraction and calculation of the flame geometry and distribution characteristics for the contour of the flame core high temperature region.