A detection algorithm for underground equipment based on feature fusion and lightweight detection head

作者：Zhongxiang Liu, Chenyang Zhou, Shichu Zhang · 年份：2026 · DOI：10.1117/12.3109860 · 研究领域：Advanced Neural Network Applications、Mineral Processing and Grinding、Belt Conveyor Systems Engineering

The underground coal mine environment is complex and fraught with multiple safety hazards. The detection of both stationary and mobile equipment is crucial for coal mine safety production and the development of intelligent mines. To address the insufficient equipment detection accuracy caused by environments characterized by low illumination, water mist, and dust, this paper proposes an improved target detection model named YOLOv11-PML, based on YOLOv11n. Its core innovations include three aspects: 1) Embedding a Parallelized Patch-Aware Attention Module (PPA) into the backbone layer to enhance feature focus on multi-scale targets; 2) Designing a Multi-Scale Context Fusion Module (MS-CFM) to replace the neck layer's feature fusion structure, thereby improving multi-scale feature fusion quality; 3) Constructing a Lightweight Shared Convolutional Detection Head (LSCDH) to reduce model size and computational cost while ensuring detection accuracy. To tackle the scarcity of underground images, SAGAN and SaMam were employed to generate a large-scale dataset simulating complex underground conditions. Experimental results demonstrate that, on the self-constructed dataset, YOLOv11-PML achieves improvements of 6.5%, 9.7%, 8.4%, 3.6%, and 2.5% over YOLOv11n in Precision, Recall, F1-Score, mAP@0.5, and mAP@0.5:0.95, respectively.