This paper presents an embedded system for real-time electrocardiogram (ECG) signal diagnosis using deep learning techniques and remote monitoring on the ThingSpeak platform. The proposed method transforms one-dimensional ECG signals into two-dimensional scalogram images, which are then classified by the SqueezeNet Convolutional Neural Network (CNN). The system classifies ECG signals into three classes: cardiac arrhythmia (ARR), congestive heart failure (CHF), and normal sinus rhythm (NSR). The PhysioNet database is utilized for training and validating the model. The classification procedure comprises signal preprocessing, image generation, model training, and performance evaluation. The trained model provides high classification accuracy, while maintaining efficient. For real-time, on-device inference, the model is deployed on an NVIDIA Jetson Nano to enable edge computing effectively. Classified results are then transmitted to the ThingSpeak cloud platform through the MQTT protocol for ongoing and remote patient monitoring.