In today’s world, high blood pressure has become a major issue for many people. The prevalence of this hypertension is increasing as a result of the consumption of certain risky foods. The purpose of our research is to classify the images of high-risk foods for hypertension patients to improve their quality of life by removing these hazardous foods from their daily eating. A dataset of 40995 food images from 15 distinct classes is used to generalize deep learning models using the transfer learning technique to fine-tune the used pre-trained models. In this case, MobileNetV2 achieved an accuracy of 95.84% across 2094 test images. Whereas Xception and VGG19 achieved accuracy rates of 81.31% and 89.97%, respectively. In comparison to other algorithms, MobileNetV2 has produced better results in less time and less misclassification. The results of the experiments illustrate that the proposed structure is capable of correctly classifying risky food images.