基于深度学习的光学薄膜元件损伤识别与分类研究
Research on damage identification and classification of optical thin film components based on deep learning
孟永 1苏俊宏 1杨国梁 1汪桂霞1
作者信息
- 1. 西安工业大学光电工程学院,西安 710021
- 折叠
摘要
激光诱导光学薄膜元件损伤是限制激光向高功率、高能量发展的瓶颈,因此,光学薄膜损伤的快速检测已成为亟需解决的问题.为提高光学薄膜元件损伤识别与分类的准确率和效率,提出了一种基于深度学习的损伤图像分类模型训练方法.采集激光辐照氧化物薄膜损伤图像,通过噪声去除、图像增强等预处理,提取损伤区域RGB值、灰度、纹理、形状等特征信息,投入BP神经网络训练识别,受数据集较少和计算误差的影响导致分类结果未达期望值,因此,使用迁移学习训练数据集,结果表明,使用迁移学习在准确率和灵敏度等评价指标上均优于BP神经网络,准确率达90％,将深度迁移学习技术用于光学薄膜元件的损伤识别,为解决激光诱导的光学薄膜损伤判别提供了新思路.
Abstract
Laser induced damage to optical thin film components is a bottleneck that limits the development of la-sers towards high-power and high-energy.Therefore,rapid detection of optical thin film damage has become an urgent problem to be solved.To improve the accuracy and efficiency of damage identification and classification for optical thin film components,a deep learning based damage image classification model training method is proposed.Collect images of laser irradiated oxide thin film damage,extract feature information such as RGB values,grayscale,texture,shape,etc.of the damaged area through preprocessing such as noise removal and image enhancement,and input BP neural network training for recognition.Due to the limited number of datasets and computational errors,the classification re-sults did not meet the expected values,Therefore,transfer learning was used to train the data set.The results showed that transfer learning was better than BP neural network in terms of accuracy and sensitivity,with an accuracy rate of 90％,The depth transfer learning technology is applied to the damage identification of optical thin film components,which provides a new idea to solve the laser induced damage identification of optical thin films.
关键词
光学薄膜/损伤检测/BP神经网络/迁移学习/分类识别Key words
optical thin film/damage detection/BP neural network/transfer learning/classification recognition引用本文复制引用
基金项目
国家自然科学基金(61378050)
国家自然科学基金(62205263)
出版年
2024
NETL
NSTL
万方数据