深度神经网络在红外图像识别中的应用  

Application of Deep Neural Networks in Infrared Image Recognition

作　　者：宗彻 ZONG Che(School of Intelligent Engineering and Automation,Beijing University of Posts and Telecommunications,Beijing 100876,China)

机构地区：[1]北京邮电大学智能工程与自动化学院,北京100876

出　　处：《河北师范大学学报(自然科学版)》2025年第3期234-242,共9页Journal of Hebei Normal University:Natural Science

基　　金：北京市自然科学基金(4132032)。

摘　　要：针对复合材料制造与使用过程中常出现的脱粘、分层等缺陷对结构安全的影响,提出一种基于深度神经网络的红外图像智能检测系统.该系统利用红外成像的高速、大面积检测优势,并构建类VOC数据集,实现对缺陷的自动判别、定位和分类.通过对比VGG,GooLeNet,ResNet及DenseNet等网络结构,最终选用ResNet嵌入Faster R-CNN框架,并采用多任务损失函数优化检测性能.实验结果表明,该方法显著提升了检测准确率与效率,为复合材料无损检测提供了高效、智能的新途径.This paper proposes an intelligent infrared image detection system based on deep neural networks to address the impact of common defects such as debonding and delamination on structural safety during the manufacturing and usage of composite materials.Leveraging the advantages of infrared imaging for high-speed and large-area inspection,the system constructs a VOC-like dataset to achieve automatic defect discrimination,localization,and classification.By comparing network architectures such as VGG,GoogLeNet,ResNet,and DenseNet,ResNet is ultimately selected and integrated into the Faster R-CNN framework,with a multi-task loss function employed to optimize detection performance.Experimental results demonstrate that the proposed method significantly improves detection accuracy and efficiency,providing a highly effective and intelligent new approach for non-destructive testing of composite materials.

关 键 词：无损检测 红外图像识别 深度神经网络 卷积神经网络 TensorFlow 

分 类 号：O175.2[理学—数学]