Coal flow rate measurement method for belt conveyors

Most existing coal flow rate calculation methods focus on contour reconstruction and volume calculation. They generally regard the bulk density of coal flow as a constant, overlooking measurement deviations caused by changes in particle-size distribution. Moreover, conventional image segmentation models provide insufficient accuracy for multi-scale coal particle segmentation in small-sample scenarios, and volume calculation methods have difficulty accurately fitting irregular coal flow contours, further increasing coal flow rate calculation errors. To address these problems, a coal flow rate measurement method for belt conveyors integrating depth-camera RGB images and point clouds was proposed from three perspectives: multiclass image segmentation, multimodal data fusion, and three-dimensional volume calculation. To segment RGB images of unit coal flow, an improved UNet3+ model was developed by embedding a Multi-Scale Attention Module (MAM) and an Adaptive Multi-Receptive Field Module (AMFM) into UNet3+ and using the H-Swish activation function, thereby achieving accurate segmentation of coal particles with multiple particle sizes. An image stitching method based on positioning-plate pixel matching was proposed to improve the stitching efficiency and registration accuracy of unit coal flow RGB images and their corresponding point clouds. The Delaunay triangulation algorithm was optimized using divide-and-conquer and adaptive methods to accurately fit three-dimensional coal flow contours and improve volume calculation accuracy. Based on the coal flow image segmentation results, area-weighted fusion was used to determine the equivalent average density of a unit coal flow, and the coal flow rate was then calculated in combination with the volume calculation result. Experimental results showed that the mean intersection over union and mean pixel accuracy of the improved UNet3+ model were 85.22% and 90.56%, respectively, representing increases of 4.71% and 3.88% over UNet3+ and 15.07% and 12.91% over UNet, respectively. Compared with the conventional constant-density calculation method, the coal flow rate measurement value obtained by the proposed method was closer to the true value.