Electrocardiographs (ECGs) are now used extensively in the diagnosis and prediction of cardiovascular diseases. Although there are several computer-aided diagnostic methods for cardiovascular disease, they all have limitations in diagnosing noisy signals. In this work, the lean model was used to overcome this restriction and identify the ECG noise zone. A deep learning model was implemented based on the Squeeze algorithm for feature analysis and the U-Net 1D technique for model compression. Each 10 s ECG signal input returns a pair of standard deviation (STD) and mean squared error (MSE) values. Next, to ascertain whether or not the signal is noisy, two thresholds were examined. In addition, the suggested solution used an optimized PanTompkins algorithm for beat detection. For the MIT BIH-DB database, our beat detection system has a sensitivity and precision rate of 99.73%, indicating its high level of precision. Additionally, our system has demonstrated notable gains in precision rates for the identification of ECG noise, especially when it comes to NST-DB records 118e_6 and 119e_6. From 66.14% and 60.66% to 95.54% and 91.51%, respectively, the precision rates for these records have grown dramatically. These outcomes verify that our technology operates with accuracy and dependability. Our proposed approach could be the foundation for further ECG signal noise detection development. It may still not effectively discriminate between clean and noisy ECG data (those with low amplitude or a strong T-wave). Consequently, to solve this problem, it is recommended that future studies optimize this model or carry out a post-processing.