基于主动学习的大规模安检图像资源库构建

Construction of large-scale Security image resource library based on active learning

文章提出了一种基于主动学习的安检图像资源库构建与优化框架。该框架采用主动学习的原理和方法,从海量的未标注的安检图像数据中筛选出最有价值的数据进行标注和学习,利用标注好的数据对数据筛选及标注中的各类模型进行训练和优化,使得框架可以不断根据资源库中已有数据调整数据筛选的标准。该框架可以在监管现场快速建立训练资源库,根据本地数据不断优化现场模型,从而提高安检工作的效率和准确性。该框架具有良好的泛化性,可扩展于地铁、海关、邮政、民航等不同的应用场景,为监管部门提供更可靠的技术支持。

This paper proposes a framework for constructing and optimizing an X-ray security inspection image resource library based on active learning.The framework uses the principles and methods of active learning to select the most valuable data from a large number of unmarked security inspection image data for annotation and learning.At the same time,the labeled data is used to train and optimize various models in data screening and annotation.The framework can continuously adjust the data selecting standards according to the data in the resource library.This framework can quickly establish a training resource library at the supervision site,continuously optimize the on-site model according to local data,improve the efficiency and accuracy of security inspection work,and reduce labor costs and error rates.The framework has good generalization and can be extended to different application scenarios such as subways,customs,postal services and aviation civil aviation,providing more reliable technical support for regulatory departments.