面向工业质量检测的云边协同增量学习方法
Cloud-Edge Collaborative Incremental Learning Method for Industrial Quality Inspection

针对计算机视觉模型在工业质量检测场景中泛化能力弱、检测精度低、推理时延高等技术瓶颈，本文提出了一种云边协同增量学习方法。该方法融合迁移学习与增量学习技术构建分层训练框架：首先采用提出的加权焦点损失函数训练云端基础模型，平衡节点间与节点内数据不平衡问题；其次在边缘节点通过冻结骨干网络的小样本迁移学习，快速适配本地业务场景得到边缘模型；设计双时间周期机制实现模型增量更新与数据上传；最终云端完成增量数据融合与模型迭代优化。该方法有效平衡了工业场景中模型性能与部署需求，为智能质量检测提供可靠技术方案。
Aiming at the technical bottlenecks of weak generalization ability, low detection accuracy, and high inference latency of computer vision models in industrial quality inspection scenarios, this paper proposes a cloud-edge collaborative incremental learning method. This method integrates transfer learning and incremental learning technologies to construct a hierarchical training framework: first, the proposed weighted focal loss function is used to train the cloud-based base model, balancing the inter-node and intra-node data imbalance issues; second, on the edge nodes, few-shot transfer learning with a frozen backbone network is adopted to quickly adapt to local business scenarios and obtain edge models; a dual-time cycle mechanism is designed to realize incremental model updates and data uploads; finally, the cloud completes incremental data fusion and model iterative optimization. This method effectively balances model performance and deployment requirements in industrial scenarios, providing a reliable technical solution for intelligent quality inspection.