Application of machine learning techniques for identifying marine species in Maltese waters

Abstract

The challenge of accurately identifying fish species was tackled in this research by employing machine learning and image classification techniques. The main aim was to develop an innovative algorithm capable of dynamically recognising the most common invasive Mediterranean fish species in Maltese coastal waters, based on available images. The target species for this study were identified as Fistularia commersonii, Lobotes surinamensis, Pomadasys incisus, Siganus luridus, and Stephanolepis diaspros. Machine learning models and transfer learning were utilised to facilitate precise, real-time species identification. The methodology involved the collection and organisation of images, followed by the training of models using consistent datasets to ensure comparable results. Among the models tested, ResNet18 was found to be the most accurate and reliable, with YOLO v8 being demonstrated as similarly effective but less consistent. These findings were highlighted to show the potential of the developed algorithm to significantly contribute to marine biology research, support citizen science initiatives, and enhance environmental management through accurate fish species identification.