面向水下光学图像的目标识别技术综述与展望

A Review and Prospects of Target Recognition Technology for Underwater Optical Images

摘要:

本文旨在对水下光学图像的目标识别核心技术范式及发展方向进行系统性综述。方法上，先明确该领域的理论基础及水下目标所具有的细粒度特征、长尾分布特性，并在此基础上重点总结样本增强、模型架构优化、迁移学习三大核心技术路径。经系统调研与综合分析，尽管该领域已在样本增强、架构优化与知识迁移等方面取得显著进展，但仍存在细粒度与长尾类别识别优化不足、复杂水下环境适配性欠缺等挑战。未来需通过构建综合性数据集、少样本学习、开放场景增量识别等前沿方向，持续提升水下目标识别的准确性与鲁棒性，为智慧海洋建设、海洋生态监测等任务提供坚实技术支撑。

Abstract:

This paper presents a systematic review of the fundamental technological paradigms and developmental trajectories for target recognition in underwater optical images. This paper first elucidates the theoretical underpinnings of the field, addressing the fine-grained features and long-tailed distribution of underwater targets. Subsequently, Three core technological pathways are summarized: sample enhancement, model architecture optimization, and transfer learning. Despite notable advancements in data enhancement, architecture optimization, and knowledge transfer, challenges persist, including inadequate optimization of fine-grained and long-tailed category identification and limited adaptability to complex underwater environments. To substantially enhance the precision and robustness of underwater target recognition, future efforts must focus on large-scale dataset construction, few-shot learning, and open-set incremental recognition. These efforts will provide robust technical support for the establishment of smart oceans and the monitoring of marine ecology.