Fish Maw Identification: Can AI Help?

Fish maw, a prized delicacy with varying prices and medicinal benefits, can be tricky to identify. Scientists recently combined two powerful tools—Wasserstein generative adversarial networks (WGAN-GP) and spectral fusion—to improve how we recognize different types of fish maw. They collected data from Raman and near-infrared (NIR) spectroscopy of four fish maw types: Beihai Male, Beihai Female, Yellow Croaker, and Red Mouth Croaker. By using WGAN-GP for data augmentation, they boosted their dataset from 300 to 3, 600 samples. They then tested three fusion strategies—data layer, feature layer, and decision layer—on two types of one-dimensional convolutional neural networks (1D-CNN), VGG and ResNet. The results? All models performed better after data augmentation. The best models were 1D-VGG (Raman)-1D-VGG (NIR) at the feature layer and 1D-ResNet (Raman)(1. 0)-1D-ResNet (NIR)(1. 0) at the decision layer, both achieving over 98% accuracy. This study shows how data enhancement and multimodal spectral data fusion can significantly improve fish maw identification, paving the way for better detection tools.