Cardiovascular illnesses can only be diagnosed and predicted with electrocardiographs (ECGs). Though portable ECG monitoring devices are available, many of them require too many resources to store and upload data to the cloud, which may be expensive and time-consuming. Our approach guarantees that dependable, clean data is submitted, saving resources and eventually leading to more precise diagnoses. Healthcare practitioners may improve patient care via more accurate diagnosis and treatment while also saving time and money by utilizing our system. In our work, applying the model which is constructed by the STM32 microcontroller with approximately 3,500 parameters, employs the U-Net 2D technique for model compression and the advanced Squeeze algorithm for feature analysis. The efficient model converts each 10 s ECG signal input into standard deviation (STD) and mean squared error (MSE) values. To detect beats, the model utilizes the improved Pan-Tompkins algorithm, and two thresholds are used to determine whether the input signal is noisy. We have demonstrated our beat detection algorithm’s extreme accuracy and dependability for the MIT BIH-DB database and the NST-DB database. Notably, our method achieves a sensitivity rate of 99.67% and a precision rate of 99.73% on the MITDB, with an efficient execution time of 42 ms. Moreover, we have seen notable gains in accuracy rates using our method to detect ECG noise, especially for NST-DB recordings 118e_6 and 119e_6, the accuracy rates for these records have grown dramatically. The methodology presented in this work could serve as a fundamental for future ECG signal noise detection research and development.