A high-confidence instance boundary regression approach and its application in coal-gangue separation

Engineering Applications of Artificial Intelligence, Год журнала: 2024, Номер 132, С. 107894 - 107894

Опубликована: Янв. 28, 2024

Язык: Английский

---

Multifractal characteristics on pore structure of grouted coal using nuclear magnetic resonance for anti-seepage evaluation

Physics of Fluids, Год журнала: 2025, Номер 37(4)

Опубликована: Апрель 1, 2025

Grouting technology is crucial for controlling rock deformation and ensuring stability in underground engineering. In this study, the grouting simulation experiments were implemented using superfine cement (SC) a water-soluble modified resin (MR) as slurries to evaluate anti-seepage effect of broken coal. The nuclear magnetic resonance (NMR) conducted analyze pore structures various grouted specimens. accordance with NMR results, multifractal characteristics types within specimens identified geometric fractal method. results showed that dimension Ds, based on seepage pore, ranged between 2 3, accompanied by high R2 values, effectively reflecting structure distribution characteristics. Specifically, Ds values MR specimen are closer porosity SC 2.12 times greater than specimens, indicating superior efficacy material. Moreover, uniaxial compressive demonstrate mechanical strength strain capacity exceed those This finding implies lower denser contribute enhanced homogeneity coal, thereby bolstering overall structural integrity. Therefore, NMR-based method can be regarded scientific feasible approach evaluating coal actual

Язык: Английский

---

A large-scale open image dataset for deep learning-enabled intelligent sorting and analyzing of raw coal

Scientific Data, Год журнала: 2025, Номер 12(1)

Опубликована: Март 8, 2025

Under the strategic objectives of carbon peaking and neutrality, energy transition driven by new quality productive forces has emerged as a central theme in China's development. Among these, intelligent sorting analysis raw coal using deep learning constitute pivotal technical process. However, progress preparation China been constrained absence accurate large-scale data. To address this gap, study introduces DsCGF, large-scale, open-source image dataset. Over past five years, extensive samples were systematically collected meticulously annotated from three representative mining regions China, resulting dataset comprising over 270,000 visible-light images. These images are at multiple levels, targeting primary categories: coal, gangue, foreign objects, designed for core computer vision tasks: classification, object detection, instance segmentation. Comprehensive evaluation results indicate that DsCGF can effectively support further research into coal.

Язык: Английский

---

Improved mask R-CNN-based instance segmentation model for coal gangue

International Journal of Coal Preparation and Utilization, Год журнала: 2025, Номер unknown, С. 1 - 19

Опубликована: Май 15, 2025

Язык: Английский

---

Research on coal and gangue segmentation based on MFCCM‐Mask R‐CNN

Energy Science & Engineering, Год журнала: 2024, Номер 12(7), С. 2958 - 2973

Опубликована: Май 30, 2024

Abstract Intelligent sorting of coal and gangue is great significance to the intelligent construction mines as well green development. In this study, we propose a segmentation method with an improved classical network Mask R‐CNN, denoted Multichannel Forward‐Linked Confusion Convolution Module (MFCCM)‐Mask R‐CNN. First, design MFCCM construct feature extraction by stacking, second, multiscale high‐resolution pyramid structure realize multipath fusion information enhance position contour target, finally, head diversity information, capture more representative unique features. Training testing models using self‐built RGB data sets, accuracy algorithm reaches 97.38%, which improvement 1.66% compared original model. Compared other Unet, Deeplab V3+, Yoloact, Yolov7, model after replacing backbone network, MFCCM‐Mask R‐CNN has higher precision recall, can accurately efficient gangue.

Язык: Английский

---

Research on accurate recognition and refuse rate calculation of coal and gangue based on thermal imaging of transporting situation

Measurement, Год журнала: 2024, Номер unknown, С. 116574 - 116574

Опубликована: Дек. 1, 2024

Язык: Английский

---

Quantification of particle size and shape of sands based on the combination of GAN and CNN

Powder Technology, Год журнала: 2024, Номер 447, С. 120122 - 120122

Опубликована: Июль 28, 2024

Язык: Английский

---

SNW YOLOv8: improving the YOLOv8 network for real-time monitoring of lump coal

Measurement Science and Technology, Год журнала: 2024, Номер 35(10), С. 105406 - 105406

Опубликована: Июль 2, 2024

Abstract Due to its large size of coal and high mining output, lump is one the hidden risks in conveyor damage. Typically, can cause jamming even damage belt during transportation process. This study proposes a novel real-time detection method for on belt. The space-to-depth Conv (SPD-Conv) module introduced into feature extraction network extract features mine’s low-resolution coal. To enhance capability model, normalization-based attention (NAM) combined adjust weight sparsity. After loss function optimization using Wise-IoU v3 (WIoU v3) module, SPD-Conv-NAM-WIoU YOLOv8 (SNW YOLO v8) model proposed. experimental results show that SNW outperforms widely used (YOLOv8) terms precision recall by 15.82% 11.71%, respectively. Significantly, speed increased 192.93 f s −1 . Compared normal models, v8 overcomes disadvantages such as being overweight, parameters are reduced only 6.04 million with small volume 12.3 MB. Meanwhile, floating point significantly reduced. Consequently, it demonstrates excellent performance, which may open up new window optimization.

Язык: Английский

---

A lightweight feature extraction backbone for gangue detection with improved Resnet50

International Journal of Coal Preparation and Utilization, Год журнала: 2024, Номер unknown, С. 1 - 18

Опубликована: Авг. 22, 2024

In response to the sizable parameters and floating point operations (FLOPs) of feature extraction backbone gangue detection algorithm, a lightweight was proposed by improving Resnet50. Initially, partial 3 × convolution used replace regular in original Resnet50 for reducing FLOPs operation. Subsequently, layers batch normalization rectified linear unit after were removed their further reduction. Finally, bottleneck attention module integrated at improve mean average precision (mAP) detection. Based on three methods, ablation comparative experiments carried out. The reduction increase mAP analyzed discussed according experimental results. It could be concluded that obtained no more than 56.36% 57.42% those popular backbones, while decrease amplitude its value 0.2%.

Язык: Английский

---

Lightweight mask R-CNN for instance segmentation and particle physical property analysis in multiphase flow

Powder Technology, Год журнала: 2024, Номер unknown, С. 120366 - 120366

Опубликована: Окт. 1, 2024

Язык: Английский

---