[论文解读] FinSight-Net:A Physics-Aware Decoupled Network with Frequency-Domain Compensation for Underwater Fish Detection in Smart Aquaculture

Jinsong Yang, Zeyuan Hu+2 more|arXiv (Cornell University)|Feb 23, 2026

Water Quality Monitoring Technologies被引用 0

一句话总结

FinSight-Net 引入物理感知的 MS-DDSP 瓶颈和 EPA-FPN 颈部，以缓解水下鱼类检测中的波长相关吸收与背散射，在参数更少的情况下实现最先进的准确度。

ABSTRACT

Underwater fish detection (UFD) is a core capability for smart aquaculture and marine ecological monitoring. While recent detectors improve accuracy by stacking feature extractors or introducing heavy attention modules, they often incur substantial computational overhead and, more importantly, neglect the physics that fundamentally limits UFD: wavelength-dependent absorption and turbidity-induced scattering significantly degrade contrast, blur fine structures, and introduce backscattering noise, leading to unreliable localization and recognition. To address these challenges, we propose FinSight-Net, an efficient and physics-aware detection framework tailored for complex aquaculture environments. FinSight-Net introduces a Multi-Scale Decoupled Dual-Stream Processing (MS-DDSP) bottleneck that explicitly targets frequency-specific information loss via heterogeneous convolutional branches, suppressing backscattering artifacts while compensating distorted biological cues through scale-aware and channel-weighted pathways. We further design an Efficient Path Aggregation FPN (EPA-FPN) as a detail-filling mechanism: it restores high-frequency spatial information typically attenuated in deep layers by establishing long-range skip connections and pruning redundant fusion routes, enabling robust detection of non-rigid fish targets under severe blur and turbidity. Extensive experiments on DeepFish, AquaFishSet, and our challenging UW-BlurredFish benchmark demonstrate that FinSight-Net achieves state-of-the-art performance. In particular, on UW-BlurredFish, FinSight-Net reaches 92.8% mAP, outperforming YOLOv11s by 4.8% while reducing parameters by 29.0%, providing a strong and lightweight solution for real-time automated monitoring in smart aquaculture.

研究动机与目标

• 解决智能养殖中由吸收和散射导致的水下图像退化问题。
• 开发一个轻量级、物理感知的检测器，使特征流解耦以恢复生物线索。
• 改进多尺度特征融合，在浑浊水中保留高频细节。

提出的方法

• 提出带有四个并行分支用于频域补偿的 Multi-Scale Decoupled Dual-Stream Processing (MS-DDSP) 瓶颈。
• 实现 Efficient Path Aggregation FPN (EPA-FPN)，具有垂直长程跳跃连接和路径剪枝。
• 使用通道级软注意力根据全局统计对分支输出进行加权。
• 骨干网络为与 EPA-FPN 及 MS-DDSP 集成的 CSPDarknet，以实现鲁棒特征提取。
• 基于 Jaffe–McGlamery 光学模型来将背散射与吸收效应分离，从而支撑设计。

实验结果

研究问题

• RQ1 一个物理感知、解耦网络是否能在水下鱼类检测中超越黑盒 CNN/Transformer 检测器？
• RQ2 轻量级的 MS-DDSP+EPA-FPN 架构在浑浊度和遮挡条件下是否能在减少参数的同时保持高精度？
• RQ3 频域补偿与细节保留的融合如何影响在浑浊水中对非刚性鱼体的定位？

主要发现

模型	参数量（M）	FLOPs（G）	DeepFish mAP50	DeepFish mAP50-95	DeepFish P	DeepFish R	AquaFishSet mAP50	AquaFishSet mAP50-95	AquaFishSet P	AquaFishSet R	UW-BlurredFish mAP50	UW-BlurredFish mAP50-95	UW-BlurredFish P	UW-BlurredFish R	UW-BlurredFish mAP50（附加）
FinSight-Net (Ours)	6.7	20.4	90.3	90.2	90.4	94.6	53.9	92.2	90.8	93.6	56.4	92.4	91.5	92.8	55.4

• FinSight-Net 在三个水下数据集（DeepFish、AquaFishSet、UW-BlurredFish）上实现了最先进的 mAP50。
• 在 UW-BlurredFish 上，FinSight-Net 达到 92.8% mAP，较 YOLOv11s 提高 4.8%，参数减少 29%。
• 消融研究表明 EPA-FPN 在减少参数的同时提升 mAP50 6.0%，而 MS-DDSP 通过物理感知分支带来显著增益。
• 整体模型在搭载 EPA-FPN 与 MS-DDSP 的情况下，在 UW-BlurredFish 上达到 53.4% mAP50，参数 6.7M，FLOPs 20.4 GFLOPs。
• 泛化测试：在 DeepFish 上训练，在未进行微调的情况下评估于 UW-BlurredFish，展示对浑浊度和光照变化的鲁棒性。