Scrap steel remelting offers a low-carbon and energy-efficient solution in steel production, enhancing the value of steel scrap recycling. However, effectively evaluating scrap steel quality remains challenging due to inefficiencies and subjectivity in human grading, which limits the scalability of production. Recent studies have employed machine vision to assist in grading, but they still struggle to overcome real-world challenges such as wind shaking, object interference, and human biases. To address these challenges, we propose layered scene recognition (LASER), a framework leveraging vision foundation models for visual intelligent recycling. This framework tackles challenges by employing spatial normalization to eliminate perspective interference and wind-induced image jitter, mask-aware recognition to mitigate densely stacked interference, and consensus estimates to correct biases. Comprehensive studies demonstrate that the proposed framework achieves 4 times the data efficiency and an 83% increase in robustness during continuous operation, attaining an accuracy of 89%, which exceeds the human recognition level of 85.5% in practical evaluations.

During industrial processing, unforeseen defects may arise in products due to uncontrollable factors. Although unsupervised methods have been successful in defect localization, the usual use of pre-trained models results in lowresolution outputs, which damages visual performance. To address this issue, we propose PyramidFlow, the first fully normalizing flow method without pre-trained models that enables high-resolution defect localization. Specifically, we propose a latent template-based defect contrastive localization paradigm to reduce intra-class variance, as the pre-trained models do. In addition, PyramidFlow utilizes pyramid-like normalizing flows for multi-scale fusing and volume normalization to help generalization. Our comprehensive studies on MVTecAD demonstrate the proposed method outperforms the comparable algorithms that do not use external priors, even achieving state-of-the-art performance in more challenging BTAD scenarios.