ROL-Net: A Robust Object Localization Network with Rep-Optimization for Precision Sensing of Chicken Freshness

Traditional shelf-life labeling often fails to accurately reflect the real-time freshness of meat products, thereby imposing potential risks to food safety and exacerbating food waste. To address this critical issue, a non-destructive chicken freshness assessment method is proposed, which integrates an intelligent indicator film with a precision-enhanced detection network (ROL-Net). Specifically, the intelligent freshness indicator film is fabricated using UiO-66-Br metal-organic frameworks (MOFs) loaded with natural anthocyanins as the colorimetric indicator, and chitosan-polyvinyl alcohol (PVA) composite as the matrix. Subsequently, to tackle the challenges of small target size, fine surface texture, and subtle color differences in the accurate quantification of indicator film discoloration, an intelligent sensing algorithm for chicken freshness based on the improved YOLOv8 is developed. This algorithm incorporates two key functional modules into the ROL-Net architecture: the residual scaling (RS) fusion module, which balances the feature energy between the ×2 and ×3 branches, stabilizes the training dynamics, and accelerates model convergence; and the output plug-and-play attention (OPA) module (SE/CBAM/Coord/LSK), which enhances channel-wise and spatial selectivity, expands the effective receptive field, while preserving the original network interface and structural topology. Experimental results demonstrate that the proposed scheme outperforms all baseline models in chicken freshness detection tasks, providing an effective, real-time, and non-destructive approach for intelligent food quality monitoring.