Scholarlink

<p>Strawberries are cultivated worldwide and favored by consumers for their rich content of vitamin C, folic acid, and antioxidants, offering anti-inflammatory and anti-cancer benefits. The soluble solids content (SSC) is a key quality attribute of strawberries, determining their sweetness and flavor intensity, while collaborating with moisture to influence overall fruit juiciness and quality. The rapid and non-destructive measurement of SSC in strawberries is of great significance to their production and postharvest management. In this research, hyperspectral imaging combined with an attention-mechanism-enhanced deep learning model (PGST, based on self-attention and Transformer) was utilized to predict the SSC of strawberries. Conventional machine learning and deep learning models were established for SSC determination. The findings demonstrated that PGST yielded outstanding results in predicting SSC, with the coefficient of determination for the training set (R²), test set (R²), and overall performance exceeding 0.95, 0.94, and baselines, respectively; the PGST model attained RMSE values of 0.511 on the training set and 0.492 on the test set. Key innovations of PGST include the self-attention mechanism for explicit visualization of wavelength importance, linking learned features to chemically meaningful spectral regions; adaptive weighting of informative bands to suppress redundancy, enhancing computational efficiency and deployment suitability; and the Transformer architecture for capturing long-range spectral dependencies beyond conventional CNN- or RNN-based models. In conclusion, integrating spectral information features with the attention mechanism of Transformer through a deep learning framework can enhance the accuracy of predicting SSC for strawberries.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Non-destructive prediction of strawberry soluble solids content using hyperspectral imaging and deeplearning model

  • 期刊论文

  • 发表日期：

    2026年2月26日

查看全文
  • Shan He,
  • Jiajia Yu,
  • Ningyuan Yang,
  • Gaoshen Cai

摘要

Strawberries are cultivated worldwide and favored by consumers for their rich content of vitamin C, folic acid, and antioxidants, offering anti-inflammatory and anti-cancer benefits. The soluble solids content (SSC) is a key quality attribute of strawberries, determining their sweetness and flavor intensity, while collaborating with moisture to influence overall fruit juiciness and quality. The rapid and non-destructive measurement of SSC in strawberries is of great significance to their production and postharvest management. In this research, hyperspectral imaging combined with an attention-mechanism-enhanced deep learning model (PGST, based on self-attention and Transformer) was utilized to predict the SSC of strawberries. Conventional machine learning and deep learning models were established for SSC determination. The findings demonstrated that PGST yielded outstanding results in predicting SSC, with the coefficient of determination for the training set (R²), test set (R²), and overall performance exceeding 0.95, 0.94, and baselines, respectively; the PGST model attained RMSE values of 0.511 on the training set and 0.492 on the test set. Key innovations of PGST include the self-attention mechanism for explicit visualization of wavelength importance, linking learned features to chemically meaningful spectral regions; adaptive weighting of informative bands to suppress redundancy, enhancing computational efficiency and deployment suitability; and the Transformer architecture for capturing long-range spectral dependencies beyond conventional CNN- or RNN-based models. In conclusion, integrating spectral information features with the attention mechanism of Transformer through a deep learning framework can enhance the accuracy of predicting SSC for strawberries.