Lightweight PPE Detection via Cross-Scale Bidirectional Feature Learning in Chemical Plants

Effective detection of personal protective equipment (PPE), such as safety helmets and vests, is critical for enforcing compliance and mitigating accident risks in high-hazard occupational settings like chemical plants. However, existing methods often exhibit suboptimal performance in complex scenarios due to insufficient multi-scale feature representation and inaccurate bounding box localization. To address these limitations, this paper proposes a lightweight detection framework based on cross-scale bidirectional feature learning. The framework integrates a Bidirectional Feature Pyramid Network (BiFPN), which employs weighted cross-layer connections to facilitate efficient and precise multi-scale feature fusion, thereby enhancing the representation of PPE at varying scales. MobileNetV4 is utilized as the backbone to ensure a balance between robust feature extraction and computational efficiency, making the model suitable for deployment in resource-constrained environments. Furthermore, the MPDIoU loss function is incorporated to refine bounding box regression by penalizing discrepancies in corner point positions, which improves localization accuracy, particularly for objects with irregular aspect ratios. Experimental results demonstrate that the proposed model outperforms the baseline, achieving improvements of 0.8% in mean mAP @ 0.5,2.7% in Precision, and 1.8% in Recall, thereby validating the effectiveness of the proposed methodology.