We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
CNNs vs. transformers: A benchmark for multi-class marine debris identification
Summary
Researchers tested different AI systems to see which one is best at automatically identifying marine debris (like plastic trash) in ocean images, comparing traditional image-recognition models against a newer approach called Vision Transformers. Both types performed impressively well (98-99% accuracy), suggesting AI could soon help scientists track ocean plastic pollution faster and more efficiently. This matters because better debris tracking helps us understand where plastic waste accumulates and spreads, information that's crucial for tackling microplastic contamination in seafood and drinking water that ultimately affects human health.
Marine debris presents serious threats to aquatic ecosystems, biodiversity, and human livelihoods, with most carrying plastic wastes. This study addresses the urgent need for effective marine debris classification through advanced deep learning models. Performance of Convolutional Neural Networks like InceptionV3, ResNet-50, and VGG-16 was evaluated and compared with that of the Vision Transformer (ViT-Base/16), a novel candidate in marine debris classification. The approach utilizes a custom dataset of debris images, and integrates robust preprocessing, feature extraction, and evaluation metrics to analyze model performance. While InceptionV3 and ResNet-50 achieved marginally higher classification accuracies (99.32% vs. 98.80% for ViT(ViT-Base/16)), this difference is small and should be interpreted with caution pending formal statistical significance testing. This work provides a significant reference for the classification of surface marine debris by fine-grained categories and suggests ViT(ViT-Base/16) as a promising alternative to the conventional architecture, warranting further investigation in larger-scale and real-world deployment settings.