This study aims to solve the problems of low detection accuracy in the quantity and particle size of fine particles during coal preparation, as well as strong background interference on small−aperture screens. Coal particle groups ranging from 0.5 mm to 3 mm were selected as the research object. A machine vision detection device was established, and image acquisition and data augmentation experiments were carried out to improve image quality. The recognition effects of different edge detection algorithms on coal particles were compared and analyzed. Finally, the particle quantity, spatial distribution and particle size were quantitatively analyzed and interpreted.

Experimental results show that the optimal image enhancement parameters are determined as brightness increased by 20%, gamma correction increased by 15%, grayscale processing increased by 15% and spatial filtering increased by 20%. Under these parameters, the peak signal−to−noise ratio (PSNR) is 31.62 dB and the structural similarity index (SSIM) is 0.91. The edge detection algorithm combining threshold segmentation, morphological operation and connected domain analysis (AMC) presents better recognition performance than other algorithms, with a recognition accuracy of 86.6%. Coal particles show a multi−center aggregation and local non−uniform distribution on the screen surface. The detected particle size distribution curve is basically consistent with the trend of real values. This study proves that the AMC edge detection algorithm is effective in identifying fine coal particles and can reliably reflect particle size distribution. The method requires no model training and features fast processing speed, which provides a practical basis for the research and development of intelligent detection equipment in the coal preparation process.