CFRD-Net: Calibrated Residual Latent Diffusion Network for Pomegranate Fruit Quality and Disease Classification

Abstract

Fruits are essential components of the global diet which plays a major role in agricultural growth and the economy.However, high-value fruits like pomegranates frequently lose the market-value due to quality deterioration, ripeness variations, surface defects, and diseases.Therefore, this research introduces the Calibrated Residual Latent Diffusion Network (CFRD-Net) for pomegranate fruit quality classification.The CFRD-Net architecture integrates Latent Conditional Diffusion learning, Residual Feature Calibration (RFC), and Residual Latent Refinement (RLR).This integration enables the learning of robust, discriminative latent representations, which captures local fine-grained texture irregularities, surface defects, and global ripeness-related color distributions.The model requires 62.34 million parameters and 11.86 GFLOPs, demonstrating a trade-off between accuracy and computational cost.CFRD-Net achieved 99.33% accuracy on the Pomegranate Fruit Quality Grading Dataset and 98.75% accuracy on the Pomegranate Fruit Diseases Dataset.These results indicate that CFRD-Net performs effectively under the evaluated benchmark settings for automated pomegranate image classification.