Interpretation of Hyperspectral Images Using Integrated Gradients to Detect Bruising in Lemons

Lemons are a popular citrus fruit known for their medicinal and nutritional properties. However, fresh lemons vulnerable to mechanical damage during transportation, with bruising being common issue. Bruising reduces the fruit’s shelf life increases risk of bacterial fungal contamination, leading economic losses. Furthermore, discoloration typically occurs after 24 h, so it is crucial detect bruised fruits promptly. This paper proposes novel method detecting in using hyperspectral imaging integrated gradients. A dataset images was captured wavelength range 400–1100 nm that were sound artificially (8 16 h bruising), three distinct classes corresponding these conditions. The divided into subsets i.e., training (70%), validation (20%), testing (10%). Spatial–spectral data analyzed 3D-convolutional neural networks: ResNetV2, PreActResNet, MobileNetV2 parameter sizes 242, 176, 9, respectively. ResNetV2 achieved highest classification accuracy 92.85%, followed by PreActResNet at 85.71% 83.33%. Our results demonstrate proposed effectively detects analyzing darker pixels images, subsequently confirming presence areas through spatial distribution accumulation. Overall, this study highlights potential gradients fruits, which could help reduce food waste