Anomaly Detection of Parasitic Plankton in Brebes Eco-Waters Using Vision-Based Autoencoder AI

The escalating impact of environmental stress on coastal ecosystems necessitates reliable, scalable tools for monitoring marine biodiversity. This study proposes an unsupervised anomaly detection framework to identify parasitic and morphologically abnormal plankton in the waters of Brebes, Indonesia. The primary aim is to develop an interpretable, vision-based system capable of detecting visual anomalies without relying on labeled anomaly data. The research integrates convolutional autoencoders for reconstructing normal plankton images, Principal Component Analysis (PCA) for feature extraction, and One-Class Support Vector Machines (OC-SVM) for classification. Monthly microscopic images were obtained from selected mangrove and aquaculture pond sites in Brebes, Central Java, using portable digital microscopy under standardized field conditions. Images that exceeded a dynamic reconstruction threshold were flagged as anomalous and validated by marine biology experts. The system achieved an F1-score of 86.1%, a precision of 85.3%, and an AUC of 0.94, demonstrating high effectiveness in distinguishing between normal and anomalous plankton. With an average inference time of 0.37 seconds per image, the system supports near real-time monitoring. These results confirm the potential of the proposed method as a low-latency, field-deployable solution for aquatic ecosystem surveillance. By integrating AI-based detection with ecological expert validation, this research offers a scalable approach for marine biodiversity assessment and establishes a foundation for future adaptive environmental monitoring systems.