AI Tool Detects and Maps Abandoned Fishing Nets to Protect Marine Life

Conservationists have developed a groundbreaking AI tool called GhostNetZero.ai to identify and locate abandoned or lost fishing nets, known as ghost nets, in the world's oceans. These nets, which amount to about 80,000 square kilometers annually, pose a significant threat to marine wildlife and contribute to plastic pollution. Detecting ghost nets underwater is notoriously challenging due to their thin and nearly invisible nature, making traditional methods ineffective. GhostNetZero.ai, a joint initiative led by WWF Germany and supported by Accenture and Microsoft’s AI for Good Lab, uses side scan sonar imaging data collected from various sources, including research institutions, governments, and offshore wind power companies. The platform leverages a Convolutional Neural Network (CNN) with a DeepLabV3 architecture and ResNet50 backbone, hosted on Microsoft Azure. This advanced AI system is capable of accurately identifying ghost nets in sonar images 94% of the time. Gabriele Dederer, WWF Germany’s ghost nets project manager, explained that every sonar image includes geolocation and metadata, allowing the AI to pinpoint the exact locations of ghost nets. Once identified, local divers and fishermen verify the AI findings and provide detailed information about the size and condition of the nets. This collaboration is crucial for planning the logistics and resources required for retrieval operations, which are complex and costly. Dederer emphasized the importance of region-specific strategies, noting that the project collaborates closely with local teams in France, Estonia, and Sweden, with plans to expand further. The verification process not only confirms the AI detections but also feeds valuable feedback back into the model, enhancing its accuracy and efficiency over time. Christian Bucher, Microsoft’s liaison to the ghost net project, highlighted the simplicity of the machine learning task despite the complexity of the imaging data. Using PyTorch libraries and NVIDIA A100 TensorCore GPUs for training and inference, the team developed a robust model that can recognize the varying patterns of ghost nets in sonar images. This high accuracy rate is a testament to the effectiveness of the approach and underscores the potential of AI in environmental conservation. The development of GhostNetZero.ai began several years ago when Dederer conceived the idea of analyzing crowdsourced sonar data. However, it was the recent collaboration with Accenture and Microsoft that brought the AI component into play, significantly advancing the project’s capabilities. The platform’s ability to process and interpret large volumes of data quickly and accurately marks a significant step forward in addressing the global problem of ghost nets. The impact of this initiative extends beyond just identifying and removing ghost nets. By reducing these underwater hazards, GhostNetZero.ai helps protect marine ecosystems and the diverse species that inhabit them. The reduction in plastic pollution also benefits the broader environment, contributing to more sustainable oceans. Industry insiders praise the innovative use of AI and cloud computing to tackle a pressing environmental issue, noting that such technological solutions could pave the way for similar initiatives in other areas of conservation. WWF Germany, founded in 1971, is part of the World Wildlife Fund, a global organization dedicated to wildlife conservation and the reduction of human impact on the environment. Accenture, a leading consulting firm, and Microsoft, a world-renowned technology company, bring significant expertise in AI and cloud computing to the project. Together, they demonstrate the potential of cross-sector collaboration in solving complex environmental challenges.