For many real-world applications, capturing patterns at diverse window scales can help to discover the different periodicity of time series. At the same time, it is helpful to gain more knowledge by analyzing time series from both time-domain and frequency-domain. This study proposes a novel method to detect distinctive patterns at variable scales in time-domain and frequency-domain of time series, and discuss its application on classification. This method integrates multiple scales, the symbolic approximation and symbolic Fourier approximation techniques to explore multi-scales and multi-domain patterns efficiently in time series. Meanwhile, statistical method is applied to select some of the most discriminative patterns for time series classification, which also can effectively reduce time complexity of the algorithm. The experiments performed on various datasets demonstrate that the proposed method has higher accuracy and better interpretability. In addition, it can be extended to multi-dimensional time series easily.