Detection of Mulberry Ripeness Stages Using Deep Learning Models

Ripeness classification is one of the most challenging tasks in postharvest management mulberry fruit. The risks microbial contamination and human error manual sorting are significant; it may result quality degradation wasting processed products. Due to advanced developments computer vision machine learning, automated became possible. This study presents results developing testing a vision-based application using convolutional neural networks (CNNs) for fruit ripening stages. To reduce training cost improve accuracy classification, transfer learning was used fine-tune CNN models. models test include DenseNet, Inception-v3, ResNet-18, ResNet-50, AlexNet. Transfer classification. AlexNet ResNet-18 exhibited best performance with 98.32% 98.65% overall classifying ripeness white black mulberries, respectively. Moreover, did not change when data sets both genotypes were mixed. able classify genotype from 600 images 2.36 min an 98.03%, which superior other architectures. It indicates that model could be precise stages mulberries horticultural products, as part system.