Underwater environments, with their rich biodiversity and unique ecosystems, present significant challenges for image annotation due to the vast volume of data and the complexity of underwater conditions. This study presents an active learning approach for underwater image annotation. The proposed approach leverages active learning techniques, specifically uncertainty sampling, to intelligently select the most informative instances for annotation, thereby reducing the overall annotation effort. The study employs two probabilistic classifiers, including Logistic Regression and Naive Bayes, to iteratively train on newly annotated data. Results indicate that the Logistic Regression model outperformed Naive Bayes, achieving high accuracy (87.12%) and AUC (0.93) highlighting its robustness in this application. The Naive Bayes classifier also demonstrated good performance, with an accuracy of 75.18% and an AUC of 0.76. This research contributes to the field of underwater image annotation by proposing an efficient active learning framework that can significantly reduce manual annotation efforts while maintaining high classification accuracy.