High-throughput and accurate detection of early-stage bruising in strawberries is essential for online quality monitoring. This study proposes an Efficient1DNet model using visible-near infrared (Vis-NIR) and short-wave infrared (SWIR) dual-band hyperspectral imaging (HSI) to discriminate bruises and visualize spatial bruise distribution. The Vis-NIR-Efficient1DNet model provided higher classification accuracy (98.41%) for different bruised strawberries than the SWIR-Efficient1DNet model (87.83%) at post-harvest. The optimal Vis-NIR-Efficient1DNet model was applied for pixel-level bruise visualization and bruised-area quantification. For early bruised strawberries with a damaged area of 7.18 ± 2.21%, the pixel-wise calculated values showed a mean absolute error (MAE) of 1.20% and a root mean squared error (RMSE) of 1.48%. Higher detection accuracy was achieved for bruised area ratios from 19.00% to 28.90%, with MAE of 0.70% and RMSE of 0.88%. These results demonstrate that dual-band HSI combined with Efficient1DNet enables rapid, non-destructive, and scalable bruise detection for real-time quality monitoring of fresh produce.
Du et al. (2026) studied this question.