AN INTRINSIC AND EXTRINSIC DEFECT-BASED NON-DESTRUCTIVE FRUIT QUALITY PREDICTION USING PolyLU-CDenseNet IN HYPERSPECTRAL IMAGE

In modern agricultural practices and food supply chains, the quality of fruits is an essential factor for tropical fruits like mango and papaya. Nevertheless, the existing works failed to focus on the prediction of both intrinsic and extrinsic defects in non-destructive fruit quality analysis utilizing Hyper-Spectral Images (HSIs). Hence, in this paper, an efficient methodology for predicting both intrinsic and extrinsic defects in non-destructive fruit quality analysis using the Polynomial Linear Unit Chroma DenseNet (PolyLU-CDenseNet) and Cumulant Generating Function-based Fuzzy Interference System (CG2FIS) approaches is proposed. Primarily, input images are gathered and further preprocessed, followed by the spatial discrimination analysis utilizing Gibbs Entropy Spectral Angle Mapper (GE-SAM). Thereafter, by using Log Gaussian Kernel-Principal Component Analysis (LGK-PCA), the spectral band optimization is performed based on the analyzed spatial discrimination. Afterward, the multi-scale segmentation is carried out. Moreover, from the optimized spectral band, chemical indices are extracted. After that, from the segmented multi-scale image, the physical features are extracted. Then, by utilizing Artificial Bernoulli Protozoa Optimization (ABPO), the features are chosen from the extracted physical features and chemical indices. Next, by using PolyLU-CDenseNet, the fruit quality prediction is done centered on the selected features. If the fruit is predicted as ripe, unripe, or overripe, then the Fruit Condition Monitoring Index (FCMI) and Structural Durability Index (SDI) are computed. Lastly, centered on the calculated value of FCMI and SDI, the fruit integrity rating is determined by utilizing the CG2FIS. The experimental analysis depicted that the proposed model obtained an accuracy of 97.90%.