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

隔爆外壳外的电缆和电气设备漏电、大功率无线电发射在金属支护和机电设备金属上感生电动势放电产生的矿井电火花，会引起瓦斯和煤尘爆炸及矿井火灾事故，因此有必要尽早感知矿井电火花。影响矿井电火花识别的主要是矿井光源，为减少矿井光源对矿井电火花图像识别的干扰，提出了一种高采样频率的矿井电火花图像识别及抗干扰方法：依据电火花的最长持续发光时间和闪光光源的最短持续发光时间，计算摄像机的采样频率，保证每次电火花出现时，电火花图像只出现在1帧图像上，且矿井光源存在时，干扰光源图像至少出现在连续2帧图像上；计算每帧图像的像素灰度和，若当前帧图像的像素灰度和与前后相邻帧图像的像素灰度和的差值均大于设定的阈值，则发出矿井电火花报警信号。试验结果表明：在无干扰光源条件下，该方法可准确识别矿井电火花图像，准确率达100%；在有日光灯、白炽灯等常亮光源干扰条件下，电火花与日光灯混合图像中电火花识别准确率达99.40%，电火花与白炽灯混合图像中电火花识别准确率达99.67%；在有闪光光源干扰条件下，电火花与闪光灯混合图像中电火花识别准确率达100%。

Leakage of electricity from cables and electrical equipment outside the explosion-proof enclosure, and mine sparks generated by high-power radio transmissions on metal supports and metal of electromechanical equipment due to induced electromotive discharges, can cause gas and coal dust explosions and mine fires. Therefore, it is necessary to detect mine electrical sparks as soon as possible. The main factor affecting the recognition of mine electric sparks is the mine light source. In order to reduce the interference of mine light sources on mine electric spark image recognition, a high sampling frequency mine electric spark image recognition and anti-interference method has been proposed. Based on the longest continuous emission time of the electric spark and the shortest continuous emission time of the flash light source, the sampling frequency of the camera is calculated to ensure that the electric spark image only appears in one frame of the image each time the electric spark appears. When the mine light source exists, the interference light source image appears on at least 2 consecutive frames of image. The method calculates the pixel grayscale sum of each image frame. If the difference between the pixel grayscale of the current frame image and the pixel grayscale sum of adjacent frames is greater than the set threshold, a mine electric spark alarm signal will be issued. The experimental results show that under the condition of no interference light source, this method can accurately recognize mine electric spark images with an accuracy rate of 100%. Under the interference of constant light sources such as fluorescent lamps and incandescent lamps, the recognition accuracy of electric sparks in mixed images of electric sparks and fluorescent lamps reaches 99.40%. The recognition accuracy of electric sparks in mixed images of electric sparks and incandescent lamps reaches 99.67%. Under the interference of a flashing light source, the accuracy of electric spark recognition in the mixed image of electric spark and flash lamp reaches 100%.

国家重点研发计划资助项目（2016YFC0801800）。

李小伟，王建业. 高采样频率的矿井电火花图像识别及抗干扰方法研究[J]. 工矿自动化，2023，49(8)：88-93, 147.

LI Xiaowei, WANG Jianye. Research on high sampling frequency mine electric spark image recognition and anti-interference methods[J]. Journal of Mine Automation，2023，49(8)：88-93, 147.