<p>Cryptographic algorithms are essential for ensuring the confidentiality, integrity, and availability of information. However, in practical applications, cryptographic algorithms can be vulnerable to side-channel attacks (SCA). Currently, side-channel attacks based on conventional convolutional neural networks (CNNs) typically rely on a single type of data for processing. This approach may result in the loss of important information and presents limitations in capturing complex patterns and deep-level features within the data, thereby affecting the effectiveness and efficiency of feature extraction. To address these issues, this paper proposes a Multi-Scale Information Fusion-based Enhanced Dual-Channel Convolutional Neural Network (MF-EDCNN) for side-channel analysis. The proposed method leverages multi-scale information fusion by integrating side-channel features from both the time domain and the frequency domain, using two parallel channels to comprehensively analyze signal characteristics. This enables more effective extraction and interpretation of side-channel information leaked during the operation of cryptographic devices, ultimately leading to accurate key prediction. Experimental results demonstrate that the proposed method achieves faster convergence on public datasets and exhibits strong performance in terms of key recovery efficiency. On the ASCAD baseline dataset, only 47 traces are required to reduce the guessing entropy to zero, while 517 traces are needed in a real deployment scenario using power traces collected from actual measurements. Meanwhile, the number of model parameters can be reduced to as low as 0.31% compared to baseline models, offering significant advantages in computational efficiency and resource consumption.</p>