Employment of Pre-trained Deep Learning Models for Date Classification: A Comparative Study

Abstract

Classification of dates in an orchard environment is a challenging task due to various texture, color, shape, and size properties. Moreover, the date has various data types that have almost the same appearance and makes classification much more difficult. To overcome these limitations, deep learning offers effective models that automatically extract features better than traditional machine learning techniques. Although deep learning models have shown excellent performance in several tasks, they require a large amount of training data to perform well. To address this issue, and to attain effective models to classify dates in an orchard environment, we employed pre-trained deep learning models. These models have been trained with a large amount of data and they showed outstanding performance in image classification. We have fine-tuned four pre-trained models; GoogleNet, ResNet-50, DenseNet and AlexNet for classifying date types. Our experimental results show that ResNet-50 has achieved the highest F1-score (98.14%) and accuracy (97.37%), compared to other models and previous methods that worked on the same dataset.