<p>Time series classification is crucial in many domains, including finance, medical treatment, and meteorology. To address the shortcomings of current approaches in adaptability to various datasets and viewpoint invariance, we propose a novel time series classification approach named Extended Capsule Networks and Time-Frequency Spectrum (EC-TFS). EC-TFS introduces a unified 2-dimensional representation of time series based on the time-frequency spectrum using the Hilbert-Huang transform. This approach supports richer feature extraction by incorporating both time and frequency-domain information. Additionally, we integrate capsule networks to enhance the detection of local features by considering their location, orientation, and relationships. Significantly, we extend the traditional capsule network to improve the representation ability of primary capsules and enhance the dynamic routing effects and efficiency to the classification capsules. Experimental results on 30 datasets demonstrate that EC-TFS improves the F1 score by 1.8% to 6.6% compared with state-of-the-art approaches. Ablation studies also reveal that the Extended Capsule Module boosts the F1 score by approximately 4.5%. Moreover, EC-TFS shows higher adaptability across various datasets compared to other approaches.</p>