Low-cost detection method for early-stage mildew in Hami melons based on hyperspectral image reconstruction using Sim-MST

文献类型： 外文期刊

第一作者： Dong, Fujia

作者： Dong, Fujia;Xu, Ying;Ma, Benxue;Zhang, Minghui;Wu, Fan;Yu, Guowei;Dang, Fumin;Li, Yujie;Ma, Benxue;Ma, Benxue

作者机构：

关键词： Hyperspectral imaging; RGB images; Spectral reconstruction; Hami melon; Sim-MST plus plus; 3D-Dilated attention convolutional neural; network

期刊名称：FOOD CONTROL （ 影响因子：6.3； 五年影响因子：6.1 ）

ISSN： 0956-7135

年卷期： 2025 年 177 卷

页码：

收录情况： SCI

摘要： Postharvest diseases induced by pathogenic microbial species account for the spoilage of fruit. The early-stage mildew detection is critical for the prevention and control of spoilage. With the rapid development of big data and deep learning, hyperspectral image (HSI) reconstruction using RGB images provides a new solution for low-cost and rapid detection of agricultural and biological products. The feasibility of RGB-to-HSI for early-stage mildew detection of Fusarium equiseti infestation in Hami melons was explored in this study. The twodimensional correlation spectroscopy was introduced to extract the sensitive spectral variables of 13 channels. Moreover, the validity of the improved Sim-MST++ and 11 state-of-the-art reconstruction algorithms was evaluated by multivariate visualization. Reconstruction consistency was evaluated using SSIM and t-SNE. The comprehensive quality attribute system (PSNR-Params-FLOPS) was proposed to determine the best reconstruction model. Furthermore, the 3D-DACNN was established for detecting early-stage mildew in Hami melons. The results showed that Sim-MST++ outperformed other reconstruction models, achieving the MRAE, RMSE, PSNR and SSIM values of 0.0446, 0.0158, 36.270 dB and 0.9333, respectively. Meanwhile, the classification model on the Sim-MST++ reconstructed data set achieved an accuracy of 92 %. These findings provided a novel method for HSI-reconstruction-based agri-food quality and safety detection.

分类号：

  • 相关文献
作者其他论文 更多>>

微信小程序