The article addresses the issue of clean water inspection for sailors at sea, highlighting two main limitations: the current clean water dataset is limited, and the classification of clean water is underdeveloped. Public datasets mostly categorize water into two levels—drinkable and non-drinkable—without providing more detailed classifications (different levels of water quality). Developing a more comprehensive dataset will take time, as data will need to be added gradually. Additionally, the dataset outputs will need to be more diverse, with multiple layers and sub-classifications to better represent water quality. To tackle this problem, the article proposes using EKI (Evolving with Klinkenberg’s Idea) algorithms, which belong to the family of incremental learning methods. These algorithms can be applied to the existing datasets despite their limitations, and with EKI’s incremental learning capabilities, the system can adapt to new requirements without needing a complete redesign, such as modifying class structures or adding more classes and sub-classes of water quality levels and categories. The article also compares four different EKI algorithms through experiments and identifies the most effective one, achieving a high level of accuracy in clean water inspection. The system will be converted to an IoT format to establish an electronic circuit for forming a water quality inspection device in the future.