AI and Sonar in Deep-Water Shipwreck Archaeology

Maritime archaeologists are revolutionizing deep-water shipwreck discovery and preservation by combining sidescan sonar anomaly classification with photogrammetric artificial intelligence. This article explores how researchers use machine learning algorithms to analyze acoustic data from the seafloor to locate hidden vessels, and how they subsequently apply computer vision to 3D models to monitor the biological and physical degradation of ancient wooden hulls over time.

Sidescan Sonar Anomaly Classification

Discovering shipwrecks in deep water requires scanning vast stretches of the ocean floor where light cannot penetrate. Archaeologists use sidescan sonar, a system towed behind a ship or mounted on an Autonomous Underwater Vehicle (AUV), which emits acoustic pulses to create an image of the seabed based on the intensity of the sound reflected back.

Historically, human experts had to manually review thousands of hours of sonar data to find shipwrecks—a slow and error-prone process. Today, artificial intelligence streamlines this through anomaly classification:

• Acoustic Shadow Analysis: AI algorithms, particularly Convolutional Neural Networks (CNNs), are trained to recognize the distinct geometric shapes and acoustic shadows cast by man-made objects protruding from the flat seabed.
• Material Differentiation: Machine learning models analyze the strength of the acoustic backscatter. Hard surfaces like wooden timbers, metal cannons, or ceramic amphorae reflect sound differently than surrounding silt, sand, or natural rock formations.
• False-Positive Reduction: By training on massive datasets of known wrecks and geological features, the AI filters out natural anomalies like boulder fields, allowing archaeologists to focus their limited diving resources on high-probability targets.

Photogrammetry in the Deep Sea

Once a shipwreck is located via sonar, archaeologists deploy Remotely Operated Vehicles (ROVs) equipped with high-definition cameras and lighting arrays to visually document the site. Because deep-water sites are inaccessible to human divers, researchers use photogrammetry to bring the wreck to the surface virtually.

The ROV flies in a tight grid pattern over the wreck, capturing thousands of overlapping two-dimensional photographs. Specialized software then processes these images, aligning common points to construct a millimeter-accurate, three-dimensional digital twin of the shipwreck.

AI-Driven Timber-Degradation Monitoring

Ancient wooden shipwrecks face constant threats from deep-sea currents, chemical erosion, and wood-boring organisms like Teredo navalis (shipworms). By applying artificial intelligence to the 3D photogrammetric models, archaeologists can monitor the health of a wreck non-invasively.

• Automated Feature Extraction: Computer vision algorithms scan the 3D models to automatically identify and map specific signs of decay, such as the tiny entry holes of wood-boring mollusks, surface splintering, and the spread of bacterial mats.
• Volumetric Change Detection: By conducting photogrammetric surveys over several years, AI compares the chronological 3D models. It calculates the exact volume of timber lost to erosion and detects micro-shifts in the hull's structural integrity that are invisible to the naked eye.
• Predictive Preservation: Machine learning models combine this degradation data with environmental metrics—such as water temperature, salinity, and current velocity—to predict how fast the wreck will deteriorate. This data helps archaeologists decide whether a site can be left alone, requires physical covering for protection, or needs immediate artifact recovery.