Eastern Canada, located in the stable continental interior of the North American Plate, experiences relatively infrequent seismic activities. Yet, historical data records significant earthquakes in seismic zones like Western Quebec (WQSZ), Charlevoix (CHZ), Lower St. Lawrence (LSZ), Southern Great Lakes (SGLZ), and Northern Appalachians (NASZ). Enhancing seismic catalogs is crucial for understanding intraplate seismicity, identifying active faults, and developing mitigation strategies in Eastern Canada. A key challenge in this sparsely monitored area is distinguishing seismic sources from industrial blast events based on phase detections from several new developed machine-learning methods. Traditional analysis methods, such as P- and S-phase amplitude/corner frequency ratio, often prove inadequate due to the complex waveforms and low signal-to-noise ratios in these intraplate settings. Our study introduces an effective discrimination method using state-of-the-art image classification convolutional neural networks (CNNs), including ResNet, VGG, ConvNeXt, and EfficientNet. We processed approximately 80,000 three-component waveforms of labeled seismic and blast events from the Natural Resource Canada (NRCan) catalog in Eastern Canada (2000-2024), converting them into spectrogram matrices similar to RGB color scale pixels. Also, incorporating the event origin time into the model further refined accuracy. Utilizing renowned image classification CNN architectures, we consistently achieved accuracies exceeding 94%. This method marks substantial progress in distinguishing between natural and anthropogenic seismic sources, proving effective even amidst low signal-to-noise scenarios. This methodology offers a straightforward and highly effective tool for seismic hazard assessment in Eastern Canada, promoting easier public adoption and contributing to preventive measures against seismic risks.