A simple and efficient sleep analysis algorithm was designed based on single-channel electrocardiogram (ECG) signals, in order to reduce the dependence on polysomnography (PSG) system. First, maximum overlap discrete wavelet transform（MODWT）was used to perform multi-resolution analysis on the original signal, then to furture extract peak information. Then, the multi-dimensional heart rate variability（HRV）features were extracted based on the first-order deviation of the peak position. To further screen the HRV features with a strong correlation with different sleep stages, a feature extraction method was proposed based on the ReliefF algorithm and Gini index. On this basis, the INFO-ABCLogitBoost method was used to mine the correlation between HRV and different sleep stages, thereby achieving a fine classification of sleep stages. Experimental results on actual public data sets showed that the proposed model had an overall accuracy of 83.67%, an accuracy rate of 82.59%, a Kappa coefficient of 77.94%, and an F1-Score value of 82.97% in the sleep staging task. Compared with conventional models in sleep staging tasks, the proposed method shows more efficient and convenient sleep quality assessment performance, which helps realize sleep monitoring in home or mobile medical scenarios.