YOLO-GBPA: Enhanced YOLO11 with Gated Bilinear Pyramid Aggregation for Dust Detection
Year: 2026
Type: conference-paper
Abstract: Dust detection in mining areas is critical for operational safety, environmental compliance, and worker health. However, detecting dust from surveillance images remains challenging because mining dust closely resembles surrounding rock, soil, and roads, while being weakly salient, diffuse, and multi-scale. Conventional YOLO-style detectors mainly rely on noise-agnostic first-order feature fusion in FPN/PAN necks. This fusion tends to blend weak dust responses with visually similar background structures, resulting in loose, misaligned, or even missing boxes on low-contrast dust. To address this problem, this paper proposes YOLO-GBPA, a dust-oriented detector that augments YOLO11s with a Gated Bilinear Pyramid Aggregation (GBPA) neck while retaining the original backbone and decoupled head. The GBPA neck performs bidirectional cross-layer spatial gating and compact hybrid first/second-order fusion between adjacent pyramid levels, enabling noise-aware, higher-order multi-scale feature aggregation. Experiments on the Dust dataset and the D-Fire fire–smoke benchmark demonstrate that YOLO-GBPA generally outperforms YOLOv5, YOLOv8, and YOLO11 baselines across most key detection metrics.
Authors: Susu Jiang, Lin Liu, Ruixin Shi, Hailong Wang
Institution: Inner Mongolia Normal University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7167487584