Comparative Analysis of Deep Learning Models for Fruit Quality Detection

In post-harvest handling, fruit quality assessment is important to increase waste minimization and aid in competitive marketing. Classical procedures of the inspection are commonly subjective, time-consuming and non-uniform. In recent years, deep learning has been showing promise in automating fruit quality inspection due to its ability to learn discriminative features from images and generalize to unseen data. In this paper, we review and compare recent developments of deep learning-based methods in fruit quality detection. The study considers different architectures (i.e., Convolutional Neural Networks, CNNs), transfer learning methods (e.g., VGGNet, ResNet, EfficientNet) and hybrid or ensemble strategies. Model accuracies, F1-scores, robustness for different scenarios are evaluated on publicly and real datasets. A summary table of results is given along with the graphical presentation, which helps for easy comparison. We also discuss the pros and cons of each model in terms of accuracy, complexity and deployment. We summarize a few main findings and future research directions in this paper, highlighting that lightweight, scalable, and interpretable deep learning methods are critical for widespread applications of DNNs in the agriculture domain.