Abstract: Sleep takes up a third of a person's life, and the quality of sleep is related to the recovery of energy in daily life, the maintenance of memory and thinking skills, concentration and other issues. Sleep assessment provides an important basis for evaluating sleep quality and preventing/diagnosing sleep disorders. Currently, Polysomnography (PSG) is the gold standard for sleep staging. However, PSG monitoring needs to attach a large number of electrodes to obtain physiological signals, which affects the sleep quality of the subjects. In clinical practice, sleep monitoring needs to be completed in the hospital and needs a sleep technician to keep vigil and sleep stage manually, which is time-consuming, labor-intensive, complex and costly. The monitoring pad based on piezoelectric film collects piezoelectric signals without requiring the subject to attach electrodes, which is an unconstrained, non-contact and non-disturbance monitoring method. It can feel the force from the thoracic motion generated by human respiration, the force caused by the heartbeat and the impact of blood flow, and the force generated by involuntary body movement during sleep. These physiological activities pass through the pad and other media to the sensor, which is converted into electrical signals. It can be placed in different positions of the bed, and can monitor physiological activities such as heart, breathing, body turning and leg movement without direct contact with the human body. Based on the characteristics of the pad that can be placed in the sleep environment for a long time, it can realize long-term and senseless sleep monitoring. So far, most researchers focus on extracting features related to heart rate, respiration and body movement from piezoelectric signals. However, in the actual daily sleep situation, the change of sleep position, body movement, and even the change of blood pressure at night will affect the strength of heartbeat information obtained by the sensor and significantly affect the continuous detection of heart rate. Meanwhile, the continuous heart rate detection throughout the night in the natural sleep state still needs to be breakthrough. This study recruited healthy young adults as subjects and used a sleep monitoring mat equipped with a piezoelectric film sensor synchronized with polysomnography (PSG) to collect data from 61 nights of sleep. Starting from signals such as cardiac activity, respiration, and body movement detected by the piezoelectric film, a total of 53 features were extracted, including 12 time-domain features, 36 frequency-domain features, and 5 higher-order statistical features. Using a bagging tree model, the study performed four-class classification (wake, N1+N2, N3, REM), three-class classification (wake, N1+N2+N3, REM), and two-class classification (wake, N1+N2+N3+REM) to predict sleep stages and validated these predictions against PSG sleep stage labels. The final testing accuracies for four-class, three-class, and two-class classifications were 80.5%, 85.3%, and 96.3% respectively, with kappa values of 0.74, 0.78, and 0.93, demonstrating excellent performance compared to similar studies. This proves the sleep monitoring mat's capability for accurate sleep assessment and monitoring, offering more possibilities for home sleep monitoring and assessment. This method also shows considerable effectiveness in sleep assessment when respiratory rate and heart rate measurements are disrupted.