Utilizing wasserstein generative adversarial networks for enhanced hyperspectral imaging: A novel approach to predict soluble sugar content in cherry tomatoes

• Published in: Food science & technology 2024-08, Vol.206, p.116585, Article 116585
• Main Authors: Cui, Jiarui, Zhang, Yao, hao, Jie, ma, Yan, Men, Jiali, pan, Shibo, Wu, Longguo
• Format: Article
• Language: English
• Subjects: Chemometrics; Cherry tomatoes; Hyperspectral imaging; Soluble sugar content; WGAN
• ISSN: 0023-6438
• DOI: 10.1016/j.lwt.2024.116585

The soluble sugar content of cherry tomatoes has a significant impact on their flavor, nutritional value, and physiological metabolism. Consequently, the development of a precise and sensitive high-spectral detection model for cherry tomato sweetness is essential for its industrial growth. However, challenges, such as the intricate nature of spectral features and low chemical content in samples, pose obstacles to this endeavor. In this study, the Wasserstein Generative Adversarial Network (WGAN) was employed to overcome these challenges and improve modeling fitting difficulties. After multiple iterations, WGAN successfully generated samples that closely resembled the original data. The results demonstrate that the performance of each model was significantly enhanced after the incorporation of the WGAN for auxiliary training. Notably, the improved dataset achieved Rc and Rp values of 0.8853 and 0.7719 for the CNN model, respectively. This study offers an innovative and efficient approach to detect soluble sugar content in cherry tomatoes and provides valuable insights into the application of artificial intelligence in the field of food and agricultural product detection. •A comprehensive spectral model for soluble sugar content in cherry tomatoes at different nitrogen levels was established.•The use of WGAN addressed challenges in creating regression models.•Detailed analysis of the spectral response of cherry tomatoes was conducted using 2D-COS.•It introduces a paradigm shift, shifting the focus from modeling methods to the study of the dataset itself.