摘要
异常检测一直是工业领域一个经久不衰的议题,传统异常检测算法往往具有可迁移能力差、操作繁琐等特点,而在深度学习领域里,图像异常检测起步较晚,但是深度神经网络也为异常检测提供了切实好用的工具并指出了新的研究方向。将自监督学习与异常检测进行融合,减少训练集对于大量的负样本数据的依赖,并比较了其他几种常见数据增强算法获取的数据集与人工标注数据集的检测效果。
Anomaly detection has been an enduring topic in industry.Traditional anomaly detection algorithms are often characterized by poor migratability and cumbersome operation,while in the field of deep learning,image anomaly detection started late,but deep neural networks also provide practical tools for anomaly detection and point to new research directions.
出处
《工业控制计算机》
2023年第11期110-111,114,共3页
Industrial Control Computer
关键词
自监督学习
异常检测
机器学习
CNN
self-supervised learning
anomaly detection
machine learning
CNN
