Jining City Jinqiao Coal Mine
CITIC Heavy Industries Kaicheng Intelligence Equipment Co., Ltd.

针对在煤矿皮带输送机上难以对煤矸石实时计量,结合图像处理技术和深度学习技术,提出基于Yolact算法的皮带矸石动态计量方法。首先对图像进行预处理,包括滤波、光照增强;然后采用轻量级的残差结构作为Yolact算法的特征提取主干,对皮带矸石进行实时的实例分割;最后二值化矸石分割结果,引入开源跨平台计算机视觉库OpenCV,利用像素阈值计算矸石的具体数量和面积,通过搭建矿用皮带矸石分拣装置,验证矸石计量算法的可行性。结果表明,皮带矸石动态计量方法能够有效学习矸石特征,网络计量皮带中矸石面积和位置信息的准确率为94.66%,网络的检测速度为30.72FPS。该方法能对煤矿皮带中的矸石进行有效计量。

In view of the difficulty for real-time measurement of coal gangue on the belt conveyor in coal mine, a dynamicmeasurement method of belt Gangue based on Yolact algorithm is proposed in combination with image processing technologyand deep learning technology. Firstly, the image is preprocessed, including filtering and illumination enhancement. Then, thelightweight residual structure is used as the feature extraction backbone of Yolact algorithm to segment the belt gangue in real time. Finally, binarize the segmentation results of gangue, introduce the open source cross platform computer vision libraryOpencv, and calculate the specific quantity and area of gangue using pixel threshold. The feasibility of the gangue meteringalgorithm is verified by building a belt gangue sorting device. The experimental results show that the proposed method caneffectively learn the characteristics of gangue. The accuracy of measuring the area and position of gangue on the belt by thenetwork is 94.% , and the detection speed of the network is 30. 72 FPS. The proposed method can effectively measure thegangue on the coal mine belt.