If not properly treated, dental caries, a common oral health issue affecting people of all ages, can lead to significant pain, various infections, and tooth decay. Traditional methods for treating dental caries rely on subjective analysis by dental professionals, which can be time-consuming, increase the risk of human errors, and potentially be expensive. To address this issue, we propose a deep learning model-based approach for the identification of dental caries. To gather a dataset of images of teeth, which includes both caries teeth and non-caries teeth, to train and evaluate deep learning-based models. We used a transfer learning method for different types of models to improve their accuracy. We tried to develop and evaluate the performance of a different deep learning model-based system using convolutional neural networks (CNN) to detect dental caries from tooth images and photographs. We examined three well-known pre-trained models: DenseNet201, ResNet50, EfficientNetB1, EfficientNetV2L, EfficientNetV2S, and MobileNet to identify the most suitable and accurate one for caries detection. The training and testing have been performed on 1554 dental caries and non-caries images. However, the MobileNet model exhibited slightly better performance when compared to ResNet50 and DenseNet201. This model obtained different percentage values: in terms of accuracy, it was 94.56%, and in terms of F1-score, it was 94%. Our study's implications extend to the development of more effective technology for early detection of dental caries.