AI in Marine Robotics & Monitoring
Toward Efficient Underwater Visual Perception through Image Enhancement, Compression, and Understanding
By Rongxin Zhu, Lei Sheng, Kaitao Wu, Azzedine Boukerche, Libo Long, Qiuling Yang
The growing demand for marine exploration, environmental monitoring, and autonomous underwater operations has elevated the role of underwater image processing. This survey addresses the fundamental constraints of harsh aquatic environments, where limited bandwidth, strong light scattering, color distortion, and complex noise degrade image quality and restrict data throughput. We review existing techniques across four core domains: image enhancement, image restoration, image compression and segmentation, and image classification. We analyze representative methods in terms of underlying principles, computational complexity, and applicability, highlighting emerging trends like deep learning, cross-modal information fusion, and resource-efficient designs for future underwater visual computing and communication systems.
Unlocking Deep-Sea Potential with AI
Our analysis reveals the transformative power of advanced AI in overcoming the challenges of underwater visual data. From enhancing clarity in turbid waters to enabling real-time object identification, these innovations drive operational efficiency and critical insights for marine exploration, environmental monitoring, and autonomous robotics.
0
Classification Accuracy
0
mAP Boost
0
UIQM Score
0
BD-Rate Reduction
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
Underwater Image Enhancement Methods
Underwater Image Restoration Methods
Underwater Image Compression and Segmentation
Underwater Image Classification
Underwater Image Enhancement Methods
Underwater image enhancement focuses on mitigating distortions caused by wavelength-dependent light absorption and scattering. Typical issues include severe color shifts, reduced visibility, and poor contrast. Enhancement techniques can be broadly classified into traditional non-physical model approaches and data-driven or physics-informed deep learning models.
Underwater Visual Perception Pipeline
Raw Imagery Acquisition
→
Enhancement & Restoration
→
Compression with ROI Coding
→
Reception and Understanding
| Reference | Technique and Key Idea | Improvement Highlights |
|---|---|---|
| Ghani et al. [56] | Contrast stretching in both RGB and HSV color spaces; histogram correction is applied to enhance color perception. | Enhances color while preserving natural blue-green lighting; maintains detail without over-amplifying contrast. |
| Paul et al. [58] | Proposed a dual histogram equalization method combined with adaptive logarithmic power functions. | Suppresses histogram spikes, improves entropy and feature similarity, and significantly boosts contrast. |
| Dong et al. [60] | Employed fusion-based enhancement by integrating RGB and LAB color spaces with normalized guided filtering. | Achieves color correction, local contrast refinement, and multi-scale detail preservation under noise. |
| Lu et al. [61] | Proposed a self-similarity-based underwater super-resolution method that merges descattering and image fusion to recover details lost to light scattering and absorption. | Preserves high-frequency details during descattering and employs convex fusion to produce sharper, color-faithful super-resolved images. |
0.94
Highest SSIM in Image Structural Fidelity
Advanced deep learning models have demonstrated exceptional capabilities in restoring structural integrity and visual quality in degraded underwater images.
Underwater Image Restoration Methods
Underwater image restoration techniques aim to recover illumination and color fidelity by reversing the degradation process introduced by light absorption and scattering. Unlike enhancement methods that emphasize perceptual improvement, restoration approaches reconstruct the scene based on optical models and prior knowledge, targeting the physical fidelity of color and structure.
Revising the Underwater Image Formation Model
Akkaynak and Treibitz [87] revolutionized restoration by identifying inaccuracies in previous models. Their revised image formation equation explicitly accounts for wavelength-dependent attenuation unique to water, validated through in situ experiments, providing a physically accurate basis for robust underwater image restoration tasks.
Impact: This foundational work enables more accurate and robust physics-based restoration algorithms, significantly improving the fidelity and interpretability of reconstructed underwater scenes. It addresses a critical gap in understanding light propagation underwater.
32.09 dB
Highest PSNR in Restoration Performance
Deep learning techniques, particularly GANs, have pushed the boundaries of image restoration, achieving impressive peak signal-to-noise ratios.
Underwater Image Compression and Segmentation
Underwater image compression and segmentation are critical for efficient data transmission and semantic analysis in resource-constrained environments. These tasks are often jointly optimized through ROI coding, where segmentation identifies key regions (e.g., marine organisms and man-made objects) and compression prioritizes these areas to preserve semantic content under limited bandwidth.
| Reference | Method Description | Technical Contributions |
|---|---|---|
| Chen et al. [99] | Proposed an adaptive compression framework based on CS to enhance robustness in UWA transmission. | Introduced a saliency-driven sampling strategy and a repair mechanism for corrupted measurements, resulting in improved perceptual fidelity. |
| Perny et al. [101] | Presented an inter-frame underwater image compression approach tailored for low-bit-rate acoustic transmission scenarios. | Demonstrated a lightweight solution for real-time image transmission in resource-constrained environments. |
| Yuan et al. [107] | Designed a human visual system (HVS)-aware compression scheme guided by underwater imaging characteristics and the Jaffe-McGlamery model. | Utilized chromaticity masking and visual masking informed by underwater color dominance and structural simplicity to enhance compression efficiency. |
3.8+ mAP
Mask R-CNN mAP Boost for Segmentation
WaterMask, an instance segmentation framework, significantly improved Mask R-CNN performance on underwater datasets by leveraging attention mechanisms and feature refinement.
Underwater Image Classification
Underwater image classification refers to the task of automatically grouping or labeling underwater visual content based on extracted visual features. It plays a pivotal role in marine life identification, habitat mapping, and environmental monitoring.
Anomaly Detection for Marine Plankton
Pu et al. [120] developed a two-phase anomaly detection framework using Wide ResNet and CKA loss. This system can distinguish between normal and auxiliary inputs, converting feature representations into scores for anomaly-aware classification.
Impact: This framework offers high sensitivity for detecting unseen plankton types, enhancing ecological monitoring capabilities and supporting critical environmental research through automated identification.
98.7%
Classification Accuracy in Object Recognition
Deep learning models have achieved remarkable accuracy in classifying underwater objects, vital for marine biology and environmental protection.
Underwater Vision ROI Calculator
Estimate the potential return on investment for integrating advanced underwater visual perception systems into your operations. Adjust parameters to see the impact on efficiency and cost savings.
Annual Cost Savings
$0
Annual Hours Reclaimed
0 hours
Your AI Implementation Roadmap
A phased approach to integrating underwater visual perception AI into your enterprise, ensuring a smooth transition and measurable impact.
Phase 1: Discovery & Strategy
Conduct a comprehensive assessment of current underwater data pipelines, identify key challenges, and define AI integration objectives. This phase includes a detailed ROI analysis and a technology compatibility study.
Phase 2: Pilot Program & Customization
Deploy a small-scale pilot leveraging pre-trained or customized AI models for specific tasks (e.g., image enhancement or object detection). Gather performance metrics and user feedback to refine the solution.
Phase 3: Full-Scale Deployment & Integration
Roll out the AI solution across all relevant underwater operations. Integrate with existing robotic platforms (AUVs/ROVs) and data management systems, ensuring seamless data flow and real-time processing capabilities.
Phase 4: Monitoring & Optimization
Establish continuous monitoring of AI system performance, accuracy, and efficiency. Implement iterative optimization cycles based on operational data and emerging requirements, including model fine-tuning and updates.
Ready to Transform Your Underwater Operations?
Schedule a personalized consultation with our AI experts to explore how these advanced visual perception capabilities can be tailored to your specific needs and strategic goals. Unlock new efficiencies and insights today.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat