基于Openmax的水下声呐图像开放集分类
Open-set Classification of Sonar Images Based on Openmax
夏梓淇 1张建磊 1王晨 2熊明磊 3谢广明4
作者信息
- 1. 南开大学 人工智能学院,天津 300350
- 2. 北京大学 软件工程国家工程研究中心,北京 100871
- 3. 博雅工道(北京)机器人科技有限公司,北京 101111
- 4. 北京大学 工学院,北京 100871
- 折叠
摘要
在人工智能技术快速发展的背景下,开放集识别作为一个新兴的问题领域受到广泛研究.本研究对FLSMDD声呐数据集进行了开放集识别任务的设计,旨在评估Openmax算法处理未知类别样本的能力,并与传统Softmax及其阈值化变体进行比较.研究通过结合残差网络和迁移学习技术,测试了不同算法在分类准确性和鲁棒性方面的表现.结果显示:Openmax算法在整体准确率上比 Softmax提高 5%,宏观 F1 参数提升了 7%,加权F1 宏观参数提高了 6%,这表明其在处理未知类别的适应性和鲁棒性方面具有显著优势.未来研究将探索优化算法设计,以进一步提高识别精度和处理效率.本研究为开放集识别技术的发展提供了有力证据,也为深度学习在更广泛的分类问题中的应用奠定了理论和实验基础.
Abstract
With the rapid development of artificial intelligence technology,open-set recognition has been widely studied as an emerging field of classification problems.In this study,an open-set recognition task for the FLSMDD sonar dataset is designed to evaluate the ability of the Openmax algorithm to handle unknown class samples.Then it is compared with traditional Softmax algorithm and its thresholded variants.By combining residual networks and transfer learning techniques,the performance of different algorithms in terms of classification accuracy and robustness is tested.The results show that the Openmax algorithm has an overall accuracy improvement of 5%compared with Softmax,a macro-F1 improvement of 7%,and a weighted Macro-F1 increase of 6%,indicating that it has significant advantages in adaptability and robustness in handling unknown categories.Future research will explore optimizing algorithm to further improve recognition accuracy and processing efficiency.This study provides strong evidence for the development of open-set recognition technology and lays a theoretical and experimental foundation for the application of deep learning in a wider range of classification problems.
关键词
水下声呐图像/深度学习/图像识别Key words
underwater sonar images/deep learning/image recognition引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据