基于多模式视觉识别技术的智能仓储盘点系统设计与开发  

Design and Development of Intelligent Warehouse Inventory System Based on Multimode Visual Recognition Technology

作　　者：孙宏斌 李鹏忠[1] Sun Hongbin;Li Pengzhong

机构地区：[1]同济大学机械与能源工程学院,上海市201804

出　　处：《工程机械》2025年第4期12-20,I0001,共10页Construction Machinery and Equipment

摘　　要：在传统离散制造企业中,由于物料种类繁多、数量庞大且摆放位置分散,仓储管理面临盘点效率低下和盘点精度不足等问题。提出一种基于智能货架和视觉识别算法的盘点系统方案,结合Spring、Spring MVC和My Batis框架,开发了基于B/S架构的智能仓储盘点系统。该系统融合视觉识别技术,在多模式情况下使用Open CV对物料和货架上的二维码进行识别和解码,同时引入C3CBAM注意力机制和标签分配SimOTA算法优化损失函数来改进YOLOv5s神经网络,以实现多物料的精确目标检测。试验结果表明,在盘点作业场景下,该系统不仅能够在盘点作业场景下基于动态帧级多数投票策略的视频流提供关键作业信息,改进后的YOLOv5s模型在测试集上达到了100%的精确率和99.84%的召回率,显著高于传统模型,每秒处理帧率大于30,满足实时性和精度需求,还能减少人工干预,从而降低成本,提升企业竞争力。In traditional discrete manufacturing enterprises,due to the wide variety,large quantity and scattered placement of materials,warehouse management faces the problems of low inventory efficiency and insufficient inventory accuracy.An inventory system solution based on intelligent shelves and visual recognition algorithms is proposed,and combined with Spring,Spring MVC and My Batis framework,an intelligent warehouse inventory system based on B/S architecture is developed.This system integrates visual recognition technology,uses Open CV in multimode scenarios to recognize and decode QR codes on materials and shelves,and introduces the C3CBAM attention mechanism and label assignment SimOTA algorithm to optimize the loss function to improve the YOLOv5v neural network in order to achieve the accurate object detection of multi-materials.The test results show that,in inventory operation scenarios,this system can provide critical operation information based on video streams of the dynamic frame-level majority voting strategy,the improved YOLOv5v model achieves an accuracy of 100%and a recall of 99.84%on the test set,which is significantly higher than the traditional model,the processing frame rate per second is greater than 30,which meets the real-time and accuracy requirements,and it can also reduce human intervention,thereby reducing costs and enhancing enterprise competitiveness.

关 键 词：多模式 智能仓储 目标检测 盘点系统 动态帧级多数投票策略 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]