Dates are revered as sacrosanct fruits in Middle Eastern culture due to their extraordinary nutritional value and rich flavour. In addition to their delectable flavour, these fruits have the potential to address health issues, thereby fostering overall wellness and health. People are at risk of selecting varieties that are inadequate for their taste preferences and culinary requirements if they are unable to distinguish between date classes during the purchasing process. This can result in disappointment with the flavour or texture. This study especially addresses the difficulties associated with accurately predicting various categories of dates. This study creates a classification system for nine distinct categories of dates by using a dataset that includes 1685 images. The classification performance of four deep learning models and a proposed composite model (MobileNetV2 + DenseNet201) is assessed. DenseNet201 archives as accuracy of (95.64%), followed by MobileNetV2 (92.41%), Xecption (90.62%) and VGG-19 (82.70%). The highest accuracy of 98.78% obtained by the proposed DateNet model among all algorithms with high precision and recall. DateNet model that has been proposed has the potential to provide a highly accurate and dependable classification of date fruit varieties.