Comparative Study of Feature Extraction Approaches for Maritime Vessel Identification in CBIR

Maritime surveillance and monitoring systems are crucial in coastal security and resource management. Vessel recognition and identification are key tasks. However, visual inspection is a costly and labour-intensive process. This study compares methods for an automated approach for vessel identification using digital image processing. The performance of classical and Machine
Learning-based feature extraction methods is evaluated and compared using a maritime vessel dataset to verify their ability to identify different vessels. The results show that BEiT-v2 achieves the highest identification performance with a mean Average Precision (mAP) of 95.05%. VGG-19 offers the best balance between accuracy (second-highest mAP) and computational cost. These findings suggest that Machine Learning methods are valuable for vessel identification, with the optimal choice depending on the specific needs of the application.

Comparative Study of Feature Extraction Approaches for Maritime Vessel Identification in CBIR (pdf)(pdf)
BLG dos Santos, AC Lorena, JEC Cruz
Encontro Nacional de Inteligência Artificial e Computacional (ENIAC), 2024

Dataset can be found in https://zenodo.org/records/14653134

more information can be found in https://jkreuz.github.io/publications/