Marine litter has become a serious environmental issue, posing significant threats to wildlife, food security, marine industries, human health, and the ecosystem at large. To curb marine litter pollution, detection and monitoring should be done as early as possible. Nonetheless, the existing methods that use people or sensors are not reliable, effective, or precise enough to manage a vast range of marine environments. This research introduces a new method for real-time identification of marine debris using an advanced deep learning computer vision algorithm called YOLOv12, which can detect and place objects in images or video streams with high speed and accuracy. We have a dataset of fifteen different forms of marine debris common in water resources, created using aerial and underwater images. The method increases the capacity of the method to deal with the complexity of the ocean environment and its diversity by introducing custom image processing layers that interpret visual data across various scales- allowing both large debris and small fragments to be detected at the same time. According to the experimental findings, YOLOv12 works well in adverse circumstances, including those with shadows, occlusions, small items, and a large number of overlapping debris, and shows high detection rates: 83.54 % precision on a standard detection threshold and 70.25 % precision across different thresholds (a good result with respect to the performance under varying levels of confidence). The tests based on quantitative and qualitative methods prove that the method can be used in real-life scenarios, autonomous systems, such as underwater robots and uncrewed aerial vehicles. To support efforts to combat marine litter pollution, preserve natural resources, and ensure the sustainable existence of the marine environment in the long term, this paper proposes an exact and cost-effective approach to monitoring marine litter. This study is suitable for detection and monitoring applications; the current model cannot reliably distinguish between debris and entangled marine organisms, making it unsafe for autonomous cleanup robots until organism-protection safeguards are implemented. Human verification remains essential for all physical intervention decisions.
扫码关注我们
求助内容:
应助结果提醒方式:
