Maturity Grading of Jujube for Industrial Applications Harnessing Deep Learning

Abstract

Jujube is one of the popular fruits that possess high nutritional components and have economic value. Grading of jujube is a post-harvest process applied in the fruit industry for the tasks like fruit quality check, fruit species identification, price labelling, edibility duration estimation, safety, etc. This research investigates the proposed customised-CNN model and two classical models (i.e., VGG16 and AlexNet) for grading jujube fruits according to their stages of maturity. Primarily, jujube of four different maturity grades was identified on the field and collected from the field manually and their images were captured through a machine vision system. Further, image pre-processing and augmentation were performed to get the training/testing-ready dataset. Finally, a customised-CNN model was deployed and grading performance was examined over the original and augmented dataset using performance metrics of precision, sensitivity, and F1-measure. Furthermore, the model's classification accuracy was compared to that of classical models, where the proposed model surmounts both the classical models. Results reveal that the proposed model attained a high grading accuracy of 99.44% and 97.53% over the augmented and original datasets respectively. Also, the computation time and training parameters count were reduced to almost one-tenth and one-third of that of the VGG16 and AlexNet models. Results advocate that the classical model could be replaced with the proposed customized-CNN model and can be further investigated for other fruits for better classification accuracy, reduced parameters and reduced computational time.

Maturity grading of jujube for industrial applications harnessing deep learning

“…The authors express their heartfelt gratitude to the Director and Dean of the Research Centre at Rajkiya Engineering College, Sonbhadra for supporting the work and providing the necessary resources for the accomplishment of this research work. The authors acknowledge the preprint version of this wok, which was previously made available on Research Square (Mahmood et al ., 2023). They thank the platform for providing a venue for early dissemination of their findings.…”