This study introduces a novel methodology for transforming ECG signals into multi-channel matrices, aimed at enhancing the analysis and classification of cardiac arrhythmias. The process begins by extracting key features from the ECG signals, including temporal intervals, frequency-based patterns, and statistical indicators. These features are then mapped across three separate channels, with each channel capturing a different dimension of the ECG signal: temporal attributes in the first channel, frequency-related characteristics in the second, and spatial and statistical features in the third. This transformation results in a multi-channel matrix that offers a comprehensive and multidimensional representation of cardiac activity. The individual channels are then fused into a single unified matrix, preserving critical information for precise classification. The fused matrix is processed through an advanced deep learning framework, which incorporates Res-Inception blocks to exploit multi-scale feature extraction and residual learning, improving the model’s ability to identify and categorize cardiac anomalies. This approach shows great promise in enhancing the accuracy and dependability of ECG-based diagnostic systems, advancing cardiac arrhythmia detection through cutting-edge signal processing and deep learning strategies.