首页 | 杂志简介 | 征稿简则 | 投稿指南 | 常见问题 | 名词解释 | 刊物订阅 | 联系我们 | English
轻量级红外刑侦图像目标识别算法
投稿时间:2023-06-01  修订日期:2023-06-12  点此下载全文
引用本文:于晓,许靖寓.轻量级红外刑侦图像目标识别算法[J].红外,2023,44(10):43~51
摘要点击次数: 322
全文下载次数: 297
作者单位E-mail
于晓 天津理工大学电气工程与自动化学院 yx_tjut@163.com 
许靖寓 天津理工大学电气工程与自动化学院  
基金项目:天津理工大学教学基金项目(YB20-05);国家自然科学基金项目(61502340);天津市自然科学基金项目(18JCQNJC01000);天津市教委科研计划项目(2018KJ133)
中文摘要:红外刑侦图像目标识别对刑事侦查具有重要意义,但刑事案件的侦破对时间和置信度要求较高。设计一种保持优异识别精度且具备较快识别速度的轻量级红外刑侦图像目标识别算法,具有十分重要的研究价值。因此借鉴生物免疫的优良特性,设计了免疫原性深度神经网络算法。该算法通过构建先天性免疫网络和适应性免疫网络来提取图像特征,然后设置免疫原性网络增强算法在处理图像特征映射时对不同通道之间优先级的调整能力,从而提高算法的精度和速度。实验结果表明,本文算法有效实现了红外刑侦图像的快速精准识别。与VGG16、VGG19、Resnet34、Resnet50、MobilenetV2等模型相比,本文算法不仅取得了99.4%的最高测试准确率,而且还具备最快的识别速度。
中文关键词:红外图像  刑侦图像  图像识别  轻量级网络
 
Lightweight Target Recognition Algorithm for Infrared Criminal Investigation Images
Abstract:Infrared image target recognition is of great significance to criminal investigation, but the resolution of criminal cases demands high requirements in terms of time and confidence coefficient. It is of great research value to design a lightweight target recognition algorithm of infrared criminal detection image to maintain excellent recognition accuracy and high recognition speed at the same time. Therefore, the excellent characteristic of biological immunity is drawn on and the immunogenic deep neural network algorithm is designed. The algorithm constructs innate immune network and adaptive immune network to extract image features. Then an immunogenic network enhancement algorithm is implemented to adjust the priority between different channels when processing image feature mapping, so as to improve the accuracy and speed of the algorithm. Experiments show that the proposed algorithm can effectively realize the fast and accurate recognition of infrared criminal detection images. When compared with VGG16, VGG19, Resnet34, Resnet50, MobilenetV2 and other models, the proposed algorithm not only achieves the highest test accuracy of 99.4%, but also has the fastest recognition speed.
keywords:infrared image  criminal investigation image  image recognition  lightweight network
查看全文  HTML  查看/发表评论  下载PDF阅读器

版权所有:《红外》编辑部

北京勤云科技发展有限公司