Reliable electrocardiogram (ECG) signal quality estimation is essential for the development of intelligent healthcare systems, particularly in wearable monitoring application. Ensuring signal fidelity is critical to prevent diagnostic errors and reduce the risk of false alarms or missed detections due to noise and motion artifacts, common challenges in wearable ECG recordings. In this study, we present a deep learning-based approach for automated ECG signal quality estimation using a fine-tuned convolutional neural network (CNN). The model is based on LiteVGG-11 architecture, originally designed for atrial fibrillation detection, and adapted through transfer learning to classify signal quality. This fine-tuning strategy allows the model to retain relevant cardiac feature representations while learning to discriminate between clean and noisy ECG segments. The model was trained and evaluated on a real-world clinical dataset collected using a wearable ECG patch device. The dataset consists of continuous recordings from two cohorts: 56 participants over 24 hours and 117 participants over a 7-day period. Ground truth annotations for signal quality were provided by clinical experts. The proposed model achieved a sensitivity of 0.8964 ± 0.0356, specificity of 0.9783 ± 0.0028, F1-score of 0.9112 ± 0.0263, area under the ROC curve (AUC) of 0.9374 ± 0.0174, and overall accuracy of 0.9597 ± 0.0041. These results demonstrate the effectiveness of deep learning- for robust ECG signal quality estimation in wearable settings. This approach has the potential to enhance the reliability of long-term cardiac monitoring and facilitate more accurate clinical interpretation and diagnosis