DETECTION AND COMPARATIVE ANALYSIS OF MUCILAGE FROM SATELLITE IMAGES USING DEEP LEARNING METHODS

Abstract

Mucilage is an environmental problem that threatens biodiversity in marine ecosystems and poses socio-economic risks. In heavily polluted areas like the Marmara Sea, early detection of mucilage is crucial for maintaining ecological balance. Early detection allows policymakers to take swift action. This study utilizes deep learning methods to detect marine mucilage using satellite imagery. The study employed YOLOv7, YOLOv8, YOLOv11, and YOLOv12 models, along with transformer-based RF-DETR and Roboflow 3.0 architectures. A dataset comprising 1113 images from various satellite sources, with mucilage regions marked with bounding boxes, was used. The dataset was expanded using data enhancement techniques. The training process was improved by applying hyperparameters to all models, resulting in performance gains. The performance of the models used in the study was evaluated using precision, recall, mAP@0.5, and mAP@0.5:0.95 metrics. Experimental results show that the YOLOv8 model achieved higher success rates than other methods. Hyperparameter settings were found to significantly impact model performance. Evaluations indicate difficulties in mucilage detection due to lowresolution images and image complexity in coastal areas. This study demonstrates the applicability of artificial intelligence technologies for monitoring environmental problems and provides a decision-support infrastructure for early-warning systems