Cardiovascular diseases pose a significant global health challenge, and electrocardiography (ECG) plays a crucial role in their detection and classification. Consequently, developing a homecare-oriented ECG diagnosis system is highly beneficial for patients to their daily lives. We present a lightweight ECG diagnosis system, utilizing state-of-the-art sensors and advanced sensing technologies to enhance the quality of healthcare. By incorporating an attention-based residual neural network (ResNet) and the Conformer model, our system improves the accuracy and efficiency of ECG signal processing, making it suitable for real-time monitoring applications in healthcare environments. To enhance the spatial and channel information of the embedded features, we investigate the use of attention-based ResNet. Additionally, we employ the Conformer neural network, which incorporates a residual mechanism, to extract both local features and global contextual information. Experimental results demonstrate that our proposed approach outperforms existing models such as wide and deep transformer neural network (denoted as PRNA), weighted ResNet, and squeeze-and-excitation ResNet. Compared with ResNet Transformer, our method is more compact in size while achieving similar performance levels. These findings indicate that our system offers a resource-efficient and high-performance solution for ECG diagnosis, making it a promising candidate for real-world healthcare applications.