Identification of Goat Milk Adulterated with Cow Milk Based on Total Synchronous Fluorescence Spectroscopy Combined with CNN

Goat milk is rich in short-chain fatty acids, which are beneficial to health; however, the adulteration of goat milk with cow milk in the market poses a significant risk to individuals allergic to cow milk. This study utilizes the differences in total synchronous fluorescence spectra (TSFS) between cow milk and goat milk, combined with convolutional neural network (CNN), to detect goat milk adulteration. An improved algorithm was introduced: a Wasserstein generative adversarial network with gradient penalty (WGAN-GP) incorporating a convolutional attention mechanism. Data filtering was performed using a combination of Mahalanobis distance and K-means algorithm. Four classical CNN classifiers—AlexNet, DenseNet121, VGG16, and ResNet50—were evaluated under consistent training conditions. Comparative analysis shows that using a 1:2 sample enhancement ratio in conjunction with AlexNet plus WGAN-GP is most effective, achieving an accuracy of 97.78% after hyperparameter optimization. This study demonstrates that integrating TSFS with CNN offers a robust method for milk fingerprint recognition.