Development of a Seafloor Litter Database and Application of Image Preprocessing Techniques for UAV-Based Detection of Seafloor Objects

Marine litter poses a significant global threat to marine ecosystems, primarily driven by poor waste management, inadequate infrastructure, and irresponsible human activities. This research investigates the application of image preprocessing techniques and deep learning algorithms for the detection of seafloor objects, specifically marine debris, using unmanned aerial vehicles (UAVs). The primary objective is to develop non-invasive methods for detecting marine litter to mitigate environmental impacts and support the health of marine ecosystems. Data was collected remotely via UAVs, resulting in a novel database of over 5000 images and 12,000 objects categorized into 31 classes, with metadata such as GPS location, wind speed, and solar parameters. Various image preprocessing methods were employed to enhance underwater object detection, with the Removal of Water Scattering (RoWS) method demonstrating superior performance. The proposed deep neural network architecture significantly improved detection precision compared to existing models. The findings indicate that appropriate databases and preprocessing methods substantially enhance the accuracy and precision of underwater object detection algorithms.