In recent years, the number of patients with hypertension has increased. Hypertension is an invisible killer. Long-term hypertension can cause a series of cardiovascular diseases such as angina pectoris, stroke, and heart failure. Therefore, early evaluation and grade assessment of blood pressure (BP) are essential to human health. The seventh report of the National Joint Committee for the Prevention, Detection, Evaluation, and Treatment of Hypertension in the United States (JNC7) classified BP levels into normotension (NT), prehypertension (PHT) and hypertension (HT). In this paper, we adopted a deep learning model (ResNet18) based on the ensemble empirical mode decomposition (EEMD) and the Hilbert Transform (HT) to predict the risk level of BP only using photoplethysmography (PPG) signals. We collected 582 data records from the Multiparameter Intelligent Monitoring in Intensive Care database (MIMIC), and each file contained arterial BP signals as the labels for inputs and the corresponding PPG signals as the inputs. Besides, the last fully connected layer of the model was initialized. We conducted three classification experiments: HT vs. NT, HT vs. PHT, and (HT + PHT) vs. NT, the F1 score of these three classification experiments is 88.03%, 70.94%, and 84.88%, respectively. A quick and accessible noninvasive BP evaluation method was offered to low- and middle- income countries.