Automated Feature Extraction and Classification of Submerged Cultural Heritage Assets in the Puck Lagoon via Multisensor Remote Sensing

ABSTRACT This study presents a strong framework for the detection and classification of Submerged Cultural Heritage Assets (SCHA) in shallow marine environments using the integration of multibeam echosounder and airborne LiDAR bathymetry with object‐based image analysis and fuzzy logic–based classification. A total of 57 remote sensing anomalies at Puck Lagoon were identified and diver‐based ground‐truthed. Of these, 42% were confirmed as cultural heritage assets, for example, a World War II aircraft crash site, demonstrating the efficiency of the technique. The high‐dimensional feature space was diminished to two components explaining 61.7% of variance with principal component analysis, which revealed clear clustering of anthropogenic and nonanthropogenic objects. Random forest feature importance ranking and ANOVA F statistic test identified vector ruggedness measure and mean LiDAR intensity as the most discriminative features for automated classification. Application of fuzzy membership functions in knowledge‐based rules modelled overlapping feature values and measurement uncertainty, improving classification reliability. This technique extends the limits of nonintrusive underwater surveying and provides an open, methodological framework for heritage managers in shallow, protected marine environments, though site‐specific recalibration is required for transferability to contexts with differing depth, energy regime, substrate type or water clarity.