Lei He

The existing methods for measuring the volume of coal mainly rely on visible light imaging, laser scanning or binocular vision. However, due to insufficient lighting, high dust concentration and the difficulty in integrating additional sensing hardware in the underground x-ray dry separation system, their application is limited. Currently, the research on thickness or volume estimation based on x-rays has not systematically addressed the lateral pixel distortion and effective penetration thickness deviation problems caused by the fan-shaped beam geometry structure, which are the main sources of volume errors in actual conveyor belt imaging systems. To solve this problem, under a fixed imaging configuration, a physically-inspired semi-empirical grayscale-thickness model was established by considering tube voltage and current, source-to-target distance, material attenuation, and beam path thickness; the unknown parameters were obtained through experimental calibration. To reduce geometric bias caused by the fan-beam imaging geometry, two correction steps are introduced: (1) a pixel-level distortion correction model for the lateral direction of the conveyor belt, and (2) a thickness component correction that converts the optical path thickness inferred from the grayscale image into the thickness component perpendicular to the conveyor belt plane. On this basis, a mapping relationship between pixel size and real geometric size is established and used for pixel-wise volume integration. Experiments conducted on the calibrated setup show that the proposed corrections substantially reduce the systematic overestimation observed in the uncorrected model. Relative to the water-displacement reference, the final corrected model keeps the volume error within approximately ±6% for the current test set, demonstrating that the proposed framework provides a practical solution for online particle-volume estimation in conveyor-based x-ray dry sorting systems without additional sensing hardware.

To address the problems of large differences in target sizes, strong background interference, difficult feature extraction for slender and irregular targets, and the challenge of balancing model complexity and accuracy in foreign object detection on coal mine conveyor belts, this paper proposed an improved detection network, DS-YOLOv9t. The model introduces a dynamic snake convolution module in the backbone network to enhance the feature modeling capability for slender and irregular foreign objects, and adopts a lightweight multi-semantic spatial-channel synergistic attention mechanism to suppress background interference and strengthen feature response in key target regions. Experimental results show that compared with the YOLOv9t baseline, DS-YOLOv9t improves mAP@0.5, mAP@0.5:0.95 and recall by 1.5%, 1.6% and 3.9% respectively, with only 2.76 M parameters and 10.8 G FLOPs. The model achieves significant improvement in detection accuracy while maintaining lightweight.

The existing LiDAR-based obstacle detection methods for autonomous rail auxiliary transportation vehicles in underground coal mines are hindered by inaccurate detection and frequent start-stops due to the wide detection range. As a result, autonomous technology for rail auxiliary vehicles remains in the testing stage and fails to meet the operational demands of coal mines. To improve detection accuracy, we propose CADNet, a deep learning-based visual detection model. To enhance the model’s robustness while maintaining a lightweight design to adapt to harsh underground environments, we have developed a feature extraction network (C2Faster), a feature fusion network (AK_FPAN), and a deformable dynamic detection head (DyHeadv3). The experimental results show that the mean average precision (mAP₅₀) of CADNet reaches 98.0%, the real-time detection speed is 112.7 FPS, the parameters are 11.6 MB, and the total computation is 13.7 GFLOPs, which is better than current mainstream target detection models. In addition, to address the problem of frequent start-stops of autonomous rail vehicles, we propose a pixel coordinate point-based region division module and an obstacle distance measurement module, respectively, and integrate them into CADNet. By defining the effective detection region of CADNet, redundant obstacle detections are avoided. Moreover, measuring the distances of detected obstacles provides a solid foundation for intelligent decision-making of rail auxiliary transportation vehicles in coal mines.

To reduce the influence of material particle size on coal gangue identification, a particle size identification method, and an adaptive image enhancement method are proposed, which can accurately identify the particle size of poorly segmented and mutually blocked materials, effectively reduce the reflection and blur of the image surface and enhance the texture details. Through the research of coal gangue images with different particle sizes, it is found that the image quality and feature curve distribution of small particle size are different from those of large particle size, and the gradient features are worse. In this paper, the accurate identification of particle size is realized using the difference in image quality and texture, and the identification rate is 99.25%. Through the image enhancement method in this paper, 33.41% of the reflection on the image surface is removed, and the average gradient is improved by 74.01%, which effectively improves the image quality and the ability to express texture information. This algorithm has high environmental adaptability, and the identification rate can reach 99.16% in moderate illumination, 98.33% in dim illumination, and 96.33% in strong illumination. This research provides a valuable idea for image processing and identification technology based on machine vision.

To improve the foreground segmentation and location accuracy of complex coal gangue images with gray histogram distribution close to the unimodal shape, a contour detection algorithm of the grayscale fluctuation matrix is proposed. The contour and non-contour pixels of coal and gangue images are investigated, and the result indicates that the gray values of the pixels around the contour exhibit the non-uniform distribution, and the gray value changes in different directions are significantly different. Accordingly, a grayscale fluctuation matrix is built by calculating the change amplitude of pixels in different directions, and multiple features are extracted from the grayscale fluctuation matrix to realize the target contour segmentation. Furthermore, the contour is optimized using the historical and future information of the contour image, thus effectively removing numerous false contours, reproducing some hidden contours and increasing segmentation accuracy. This method has high accuracy, and the maximum error rates of the pixel area and center coordinate of contour detection are 4.404% and 3.18% respectively. This study provides a feasible solution to the edge detection and segmentation of images with similar and complex backgrounds.

Currently, new sorting technology of coal and gangue not merely focuses on the identification of coal and gangue, but also emphasizes the fine identification of coal. Therefore, this paper takes the main coal types, including one-third coking coal, gas coal, and gangue in Huainan coalfield as the research object, and proposes a method to identify coal types and gangue. The ADXRD patterns and EDXRD patterns of the samples were obtained through X-ray diffractometer and EDXRD test platform, respectively. The differences between the phase composition and the internal connection between the ADXRD pattern and EDXRD pattern of the samples were discussed, and the conversion of ADXRD pattern to EDXRD pattern of coal and gangue was simulated, the PSO-SVM identification model of coal types and gangue was established. The experimental results show that this method provides a new and effective technical way for coal types and gangue identification, which has important reference significance for fine identification of coal and gangue.

Coal and gangue separation technology based on X-ray has broad development prospects due to its low energy consumption and pollution-free environment, which is also the key part in the green and intelligent mines. Aiming at the low recognition accuracy for coal and gangue with the particle size of 5–15 mm for the dual-energy X-ray coal and gangue separation technology, the coal and gangue recognition method based on the combination of X-ray transmission and diffraction principle was proposed. The dual-energy X-ray images based on the X-ray transmission principle were collected for coal and gangue with particle size larger than 15 mm, and Rc, Glc, Gl, Ra were extracted as the recognition features. EDXRD patterns based on the X-ray diffraction principle were collected for coal and gangue with particle size less than 15 mm, and the characteristic diffraction peaks were extracted as the recognition features. Then, the PSO-SVM model was established for coal and gangue recognition. The test results show that the proposed method can broaden the particle size range for dry coal preparation based on X-ray, and the recognition accuracy of coal and gangue with particle size less than 15 mm is 98%, which is 16.7% higher than that of the method based on the X-ray transmission principle alone. The comprehensive recognition accuracy of coal and gangue with particle size of 5–100 mm reached 97.5%. Consequently, this paper provides a new technical approach for coal and gangue identification.

Accurate segmentation of dual-energy X-ray transmission (DE-XRT) coal and gangue image regions are a prerequisite for feature extraction, identification, localization, and separation. A watershed algorithm based on multi-grayscale threshold segmentation (MGTS) is proposed to mark the foreground...

In lump coal and gangue separation based on photoelectric technology, the prerequisite of using a dual-energy X-ray to locate and identify coal and gangue is to obtain the independent target area. However, with the increase in the input of the sorting system, the actual collected images had adhesion and overlapping targets. This paper proposes a pit point detection and segmentation algorithm to solve the problem of overlapping and adhesion targets. The adhesion forms are divided into open and closed-loop adhesion (OLA and CLA). Then, an open- and closed-loop crossing algorithm (OLCA and CLCA) is proposed. We used the conjugate lines to detect the pit and judge the position and distance of the pixel point relative to the conjugate lines. Then, we set the constraint of the distance of the pixel point and the relatively straight line position to complete the pit detection. Finally, the minimum distance search method was used to obtain the dividing line corresponding to the pit to complete the image segmentation. The experiment results demonstrate that the segmentation accuracy of the overlapping target was 90.73%, and the acceptable segmentation accuracy was 94.15%.