School of Artificial Intelligence, China University of Mining and Technology-Beijing

尽早发现矿井电火花并报警，可避免或减少瓦斯和煤尘爆炸及矿井火灾事故发生。井下没有日光、月光及星光等自然光源，影响矿井电火花识别的主要是矿井光源。通过调整摄像机的安装位置及角度，可以避免或减少矿井固定光源对电火花识别的影响，但不能解决矿井移动光源对电火花识别的影响。不同形式电路产生的电火花放电周期不同，但电火花放电时间均小于4 ms。闪光光源最小亮持续时间为240 ms。因此，可利用电火花发光时间短、矿井移动光源对摄像机照射时间较长的特点，排除矿井移动光源对摄像机照射的影响。摄像机在较高帧频下拍摄，电火花图像具有1帧暗-1帧亮-1帧暗的特征，即“暗-亮-暗”的帧特征。有电火花的“亮”帧，单帧图像像素灰度和较大；无电火花的“暗”帧，单帧图像像素灰度和较小。移动光源对摄像头的照射是变化的，经历了无光-有光-无光过程。在无电火花的情况下，摄像机同样在较高帧频下拍摄，无论是移动常亮光源还是移动闪光光源的图像均不会出现“暗-亮-暗”的帧特征。基于电火花图像独有的“暗-亮-暗”的帧特征，提出了基于图像邻帧像素灰度和的矿井电火花识别及报警方法：实时采集监控区域视频图像；根据设定的帧频，对视频图像进行分帧预处理，分别计算单帧图像像素灰度和；若当前帧图像像素灰度和与前一帧图像像素灰度和之差小于预先设定的阈值，则继续采集视频图像，否则计算当前帧图像像素灰度和与后一帧图像像素灰度和之差；若该差值小于预先设定的阈值，则继续采集视频图像，否则发出矿井电火花报警信号；矿井电火花报警后，若人工没有启动应急响应，则继续进行矿井电火花报警，否则退出当前报警状态，继续采集视频图像。通过该方法可有效消除移动常亮光源和闪光光源的干扰。

Early detection of mine electric sparks and alarm can prevent or reduce gas and coal dust explosions and mine fire accidents. There are no natural light sources such as sunlight, moonlight, and starlight underground. The main factor affecting the recognition of mine electric sparks is the mine light source. By adjusting the installation position and angle of the camera, the impact of fixed mine light sources on electric spark recognition can be avoided or reduced. But it cannot solve the impact of mobile mine light sources on electric spark recognition. The discharge cycle of electric sparks generated by different forms of circuits is different, but the discharge time of electric sparks is less than 4 ms. The minimum bright duration of the flash light source is 240 ms. Therefore, the features of the short emission time of electric sparks and longer exposure time of mine moving light sources to cameras can be utilized to eliminate the impact of mine moving light sources on camera exposure. The camera shoots at a high frame rate, and the electric spark image has a feature of 1 frame dark -1 frame bright -1 frame dark, that is, a "dark light dark" frame feature. The "bright" frames with sparks have a large sum of pixel grayscales in a single frame. The "dark" frames without sparks have a small sum of pixel grayscales in a single frame. The illumination of a moving light source on the camera is variable, going through a process of no light, light, and no light. In the absence of electric sparks, the camera also shoots at a high frame rate. The images of both moving constant bright light sources and moving flashing light sources do not exhibit the "dark bright dark" frame feature. Based on the unique "dark bright dark" frame feature of electric spark images, a mine electric spark recognition and alarm method based on the sum of adjacent frame pixel grayscale is proposed. The method collects monitoring area video images in real time. According to the set frame rate, the method preprocesses the video image into frames and calculates the pixel grayscale of a single frame image separately. If the difference between the current frame image pixel grayscale and the previous frame image pixel grayscale is less than the pre-set threshold, the method continues to collect the video image. Otherwise, the method calculates the difference between the current frame image pixel grayscale and the subsequent frame image pixel grayscale. If the difference is less than the pre-set threshold, the method continues to collect video images. Otherwise, the method issues a mine electric spark alarm signal. After the mine electric spark alarm, if the emergency response is not activated manually, the mine electric spark alarm will continue. Otherwise, the method exits the current alarm state and continues to collect video images. This method can effectively eliminate the interference of moving constant light sources and flashing light sources.

国家重点研发计划资助项目（2016YFC0801800）；

0 引言
1 电火花的最大放电周期及时间
2 矿井光源对电火花识别的影响
2.1 闪光光源最高工作频率
2.2 移动光源最大移动速度
2.3 矿井光源对电火花识别的影响
3 基于图像邻帧像素灰度和的矿井电火花识别及
4 结论

孙继平,李小伟,王建业.基于图像邻帧像素灰度和的矿井电火花识别及报警方法研究[J].工矿自动化,2023,49(07):1-5.DOI:10.13272/j.issn.1671-251x.18141.

SUN Jiping, LI Xiaowei, WANG Jianye. Research on mine electric spark recognition and alarm method based on the sum of adjacent frame pixel grayscale of images[J]. Journal of Mine Automation，2023，49(7)：1-5.