Date fruit type classification using convolutional neural networks

Classification of objects is an important task for convolutional neural networks (CNNs). They have been applied to numerous fields with excellent results. In this study, we use CNNs to classify five categories of Sukkari dates, namely Galaxy, Mufattal, Nagad, Qishr, and Ruttab. Transfer learning is when a pretrained model is taken and only the final layers are trained to make a prediction. In this paper, we used the following five models: SqueezeNet, GoogLeNet, EfficientNet-b0, ShuffleNet, and MobileNet V2. The results show that SqueezeNet outperforms the other networks with a classification accuracy of 92% on the testing set. The testing accuracy for GoogLeNet, EfficientNet-b0, ShuffleNet, and MobileNet V2, on the other hand are 85.14%, 82.86%, 89.14%, and 87.43%, respectively. As this is a classification task, other metrics like precision, recall, and F1 score are also evaluated. These values for the SqueezeNet on the testing set are 92.67%, 92%, and 92.33%, respectively. ShuffleNet was second with values of 89.41%, 89.14%, and 89.28%, respectively. EfficientNet scored the lowest with 83.10%, 82.86%, and 82.98%, respectively.