Container defect detection using enhanced YOLO with uncertainty-aware dynamic modulated convolution and parallelized patch-aware attention

Container defect detection is a critical task in industrial settings to ensure logistics safety and cargo integrity. However, conventional detection methods often fall short due to the diverse morphologies, varying scales of defects, and complex background interference. To address these challenges, we propose an enhanced single-stage object detection model, termed UPAD-YOLO. Specifically, we introduce a novel Uncertainty-Aware Dynamic Modulated Convolution (UDMConv), which dynamically adjusts the receptive field of convolutional kernels to efficiently extract multi-scale features, thereby accommodating the morphological diversity of defects while enhancing computational efficiency. We further integrate a Parallelized Patch-Aware Attention (PPA) module that leverages multi-branch architecture, patch-aware attention mechanisms, and both channel and spatial attention to substantially improve detection accuracy and robustness. In addition, the detection head of the original model is combined with an Adaptively Spatial Feature Fusion (ASFF) module to effectively integrate defect features across different resolution levels, thereby enhancing the model’s capability to recognize multi-scale defect patterns. Experimental results demonstrate that UPAD-YOLO outperforms the baseline model YOLOv11 in container defect detection, achieving a 4% improvement in mean Average Precision (mAP) and a 2.9% increase in F1-score, highlighting its robustness and practical applicability. This study not only offers an efficient and reliable solution for container defect detection but also provides valuable insights for broader applications in industrial visual inspection tasks.