Hunting the Invisible: How AI and Satellite Radar Are Spotting India’s Dark Ships

By: Deepak Kumar (IIT Kharagpur)

Key Highlights

India’s maritime security is threatened by thousands of small, unregistered, and AIS-off vessels that evade traditional monitoring systems.
Conventional methods and generic AI models struggle to detect small, low-contrast boats that dominate India’s fishing fleet.
SAR imagery provides all-weather, day-night coverage, but needs specialized AI to detect small vessels effectively.
Our SAR-specific AI solution bridges this gap by detecting suspicious ships even in AIS-dark zones using open satellite data.
This empowers India’s maritime agencies to protect coastal security, enforce regulations, and strengthen initiatives like Maritime Domain Awareness and the Blue Economy.

India is surrounded by water on three sides. India’s coastline stretches approximately 11,099 km, with an Exclusive Economic Zone (EEZ) of around 2.3 million km², and coastal and maritime surveillance requirements have expanded dramatically. Given the rise in untracked, AIS-off vessels involved in illegal fishing and environmental risks, safeguarding this maritime expanse is a necessity. The ocean is India’s lifeline, from coastal fishing and trade to national defense and marine biodiversity.

Yet, a major security concern continues to slip past the radar—“dark ships”, or vessels that sail without broadcasting their identity via AIS (Automatic Identification System). These vessels are invisible to conventional monitoring systems, making them a serious threat.

Our project explores how machine learning combined with satellite-based Synthetic Aperture Radar (SAR) can fill the surveillance gap and help India detect what was previously hidden at sea.

What Are Dark Ships—and Why They Matter?

Dark ships are vessels that either do not carry an AIS device or intentionally switch it off to avoid detection. AIS is designed to transmit a ship’s ID, position, speed, and heading to nearby ships and coastal authorities.

But AIS is not foolproof. It’s voluntary for small fishing boats and easily disabled on larger vessels. These gaps in visibility create dangerous loopholes. Ships that don’t broadcast AIS are essentially invisible to monitoring authorities. And in regions with dense fishing activity, like India’s east coast or the Arabian Sea, the invisibility comes with consequences.

Illegal fishing in India’s EEZ is already a major concern. Reports show that over 40% of fish stocks are either fully exploited or overfished (FAO, 2022). Meanwhile, the National Crime Records Bureau has documented multiple instances of smuggling through fishing boats, many of which operate without transponders

India’s Maritime Blind Spot: What the Numbers Say?

India continues to face a large-scale gap in AIS coverage, particularly among small vessels.

Here’s what the data reveals:

● India has over 2.5 lakh registered fishing boats (Ministry of Fisheries, 2022)

● Nearly half of India’s fishing boats still lack Automatic Identification System (AIS) or Vessel Monitoring System (VMS) devices

● This gap is especially profound among smaller, traditional vessels, which make up a significant portion of the active fleet, leaving many boats untracked.

This means that hundreds of thousands of vessels operating daily in Indian waters are completely untraceable by standard maritime monitoring systems. This invisibility creates a “blind spot” that is exploited for illegal fishing, smuggling, and other unauthorized activities, while also complicating rescue operations and maritime safety efforts.

“India’s coastline is patrolled by technology, but vast stretches remain blind to the movement of thousands of small fishing boats—creating a critical gap that fuels illegal fishing, threatens marine ecosystems and resources, and weakens maritime security.”

Why Existing Solutions Fall Short?

Multiple technologies monitor India’s coastline, but each has critical blind spots—especially for small, unregistered fishing boats:

● AIS Tracking: Depends entirely on vessels choosing to broadcast their position. Easily disabled or spoofed.

● Coastal Surveillance Radars (CSRN): Effective within ~50 km of the shoreline. Not useful in deep-sea zones.

● Optical Satellites (Sentinel-2, Planet): High-resolution but ineffective during cloudy weather or at night less impactful in Indian monsoon zones.

● SAR Data (Sentinel-1): Can see through clouds and at night, but raw SAR images are complex and noisy, requiring advanced preprocessing and calibration to identify ships.

● Traditional Algorithms (CFAR): Fail in coastal clutter, struggle to detect small and low-reflective vessels like fiberglass fishing boats.

● Generic Machine Learning Models: YOLO or Faster R-CNN and other structured models perform well on large ships but often struggle with small vessels due to feature loss, noisy sea backgrounds, and poor scale handling—especially in complex coastal regions. These models are fast and powerful but not tuned for the subtle signatures of small, AIS-off ships.

"Traditional SAR ship detection methods—like thresholding and classic algorithms—were designed for large, metal vessels and open seas, but they often miss small, low-reflectivity boats in cluttered coastal environments, leaving significant gaps in maritime surveillance and security."

The SAR-based AI Solution: Making the Invisible Visible

Synthetic Aperture Radar (SAR) imagery offers a unique advantage—it captures Earth’s surface regardless of light or weather, making it ideal for maritime surveillance. But SAR images are grainy and complex, especially around coastlines. Identifying a small vessel in a noisy sea background is like “finding a needle in a foggy, choppy haystack.” This is where machine learning enters. By training a model on thousands of annotated SAR images, the system learns to identify patterns, shapes, and brightness variations associated with ship targets—even small, low-contrast ones. Our solution utilizes C-band SAR data from Sentinel-1, providing regular imagery over India’s coastal waters. We preprocess these images into 800×800 tiles, filtering noise and normalizing contrast. Then, we apply a deep learning model optimized specifically for SAR ship detection.

The model: custom-built to spot what others miss

We use a customized version of YOLOv8n, a popular deep learning model for object detection. But we didn’t just apply it out of the box. Instead, we tailored the model for SAR-based maritime detection by integrating specialized modules, each playing a vital role:

● SR Module: Helps the model extract richer features from the SAR image by using multi-path filters. This increases detection clarity, even for blurred or distorted shapes.

● SPD Module: Keeps the fine-grained details that matter—especially for tiny boats under 20 meters, which most models ignore.

● Hybrid Attention Module: Acts like a radar operator’s focus—it filters out sea clutter and enhances real ship signals.

● Shape-NWD Loss Function: A mathematical formula that improves how well the model draws boxes around ships—important when the vessel is small or partially visible.

“In a nutshell, traditional detection models miss what matters most—small, silent vessels. Our SAR-optimized architecture brings clarity to cluttered seas, offering a smarter, sharper way to detect dark ships in India’s maritime zones.”

Why This Matters for India?

India faces a unique mix of challenges—long coastlines, thousands of unregistered boats, and a high risk of illegal activity at sea. Our solution offers a pathway to strengthen maritime surveillance using open satellite data, without depending solely on AIS or expensive commercial sensors. It can flag suspicious vessels in areas where AIS data is absent. It can assist Coast Guard, Navy, and marine enforcement teams in identifying priority zones.

This system supports Underwater Domain Awareness (UDA) by helping identify and prevent illegal fishing activities that damage marine ecosystems, disturb underwater biodiversity, and contribute to long-term environmental degradation. It strengthens UDA’s environmental protection and sustainability objectives through actionable surface-level intelligence.

It also contributes to India’s Maritime Domain Awareness (MDA) strategy, aligning with broader national initiatives like SAGARMALA and the Blue Economy by enhancing coastal surveillance and improving surface vessel monitoring across the Indian Ocean Region.

India’s maritime future depends on the ability to detect what others overlook. Our AI-powered system transforms blind spots into zones of clarity—empowering authorities to counter illegal activities and safeguard the blue economy. With every dark vessel detected and every anomaly flagged, we take a step toward a smarter, safer, and more sustainable Indian Ocean—where technology and vigilance work together to protect national interests and marine ecosystems.

Looking Forward

The current system works effectively on preprocessed SAR imagery and processed tiles. Moving ahead:

● Integrate AIS data to prioritize regions of interest and guide SAR image selection, enabling more targeted and efficient dark ship detection

● Build a user-friendly interface or dashboard to visualize detections and support decision-making for maritime authorities

● Explore deployment on real-time tasking platforms for operational surveillance use

● Work toward a lightweight deployment suitable for use on edge devices or coastal systems

References

[1] SAR Small Ship Detection Based on Enhanced YOLO Network

[2] How India ‘added’ more than 3,500 km to its coastline

Deepak Kumar

About Author

Deepak Kumar is pursuing a dual degree (B.Tech and M.Tech) in Civil Engineering from IIT Kharagpur. He is passionate about artificial intelligence, data science. He enjoys building robust, deployable systems that solve practical problems and contribute to smarter, safer environments.