Answers are grounded in this paper and cite their sources.
Rapid and accurate detection of damaged peppers is essential for quality assessment and intelligent sorting in modern agricultural production. This study developed a hyperspectral imaging–based framework for discriminating healthy and damaged green peppers using spectral information acquired within the 372.6–1039.6 nm range. Candidate wavelengths were first selected using UVE‐SPA. To identify informative band‐ratio features, a comprehensive evaluation framework integrating statistical significance and spatial discriminability was proposed. Based on this framework, three optimal band‐ratio combinations were selected to construct a three‐channel BR image, enabling enhanced lesion contrast and structural feature representation while reducing the influence of illumination variation. Independently, the MDSR method was employed to select three representative wavelengths for constructing a complementary three‐channel spectral image. To further exploit the complementary characteristics of visible and hyperspectral information, a classification‐driven fusion strategy was developed to integrate RGB images with BR and MDSR‐derived images, respectively. The resulting fusion images effectively combined color, spectral, and spatial information associated with pepper deterioration. Comparative experiments demonstrated that image fusion substantially improved feature representation and classification performance compared with single‐source image inputs. Among the evaluated models, ResNet18–MLP achieved the best performance, reaching an overall classification accuracy of 93.5%. The results indicate that the proposed band‐ratio evaluation framework and classification‐driven fusion strategy effectively enhance damage discrimination and provide a potentially effective approach for hyperspectral image‐based defect detection in peppers. This study offers a promising foundation for intelligent postharvest inspection and sorting of agricultural products.
No takes yet. Share an insight, caveat, or question.
Li et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: