Boundary-Aware Semantic Segmentation Method for Identification of Maceral Components of Deep Coal Reservoir

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Accurate identification of coal macerals provides an important basis for coal quality evaluation, reservoir heterogeneity characterization, and quantitative maceral analysis in deep coal reservoirs. Traditional manual maceral identification and point-counting methods depend heavily on operator experience and are affected by sample preparation, petrographic image quality, maceral boundary clarity, and subjective interpretation. Consequently, the efficiency and consistency of maceral identification remain limited for coal petrographic images with grayscale overlap, blurred boundaries, complex textures, and strong background interference. This study proposed a CBAM-DeepLabV3+ semantic segmentation method with boundary supervision to address these problems. Experiments were conducted on 277 reflected-light microscopic images of deep coal samples from the Ordos Basin. The images were classified into three categories: background, vitrinite, and inertinite. Under the same data partitioning, training strategy, and evaluation metrics, DeepLabV3+, CBAM-DeepLabV3+, and the proposed model were compared through ablation experiments. On the independent test subset, the developed model achieved a PA of 0.9040 and an mIoU of 0.8406. Compared with the baseline DeepLabV3+, the mIoU increased by 0.0221; compared with CBAM-DeepLabV3+, Boundary F1 and Boundary IoU increased by 0.0252 and 0.0227, respectively. Performance comparisons showed that the proposed method better preserved boundary continuity and reduced local misclassification in regions with blurred boundaries, pores and fractures, mineral background interference, and complex fine-grained inertinite. These results support automated maceral identification and quantitative characterization of maceral spatial distribution in deep coal reservoirs.