According to the World Health Organization, almost 17 million people die each year as arnresult of cardiovascular illness. The irregularity and abnormalities of heartbeat rhythm whichrnis known as arrhythmia is one of the conditions that can affect the cardiovascular system.rnElectrocardiogram (ECG) is a reliable tool that can be used for monitoring the cardiovascularrnhealth. Recently, classifying the ECG signals based on Artificial Intelligence (AI) isrnincreasingly being studied. Convolutional Neural Networks (CNN) in particular have beenrneffectively applied for the classification of ECG signals. Although high prediction accuraciesrnhave been reported, majority of previous studies have only been developed to classify limitedrnnumber of arrhythmias. The methods were developed to evaluate all major types ofrnarrhythmias using 1-D CNN to classify time domain representation of ECG waveforms.rnHowever, using 1-D CNNs has limited flexibility due to the use of 1-D kernels. There arernmethods reported to transform the time series signals into 2-D images using STFT and use 2-rnD CNN. However, STFT is difficult to apply to non stationary signals; there is no way tornresolve the complete frequency content of such signals with a single localizing window size.rnTo overcome this obstacle of Fourier decomposition, the Continuous Wavelet Transformrn(CWT) could be used to breakdown a signal into wavelets with a high degree of temporalrnlocalization. The S-transform could be another option since it takes the advantage of STFTrnand wavelet. This thesis study uses CNN classifiers for detecting and classifying heartrnarrhythmias based on analysis of ECG signals in time-frequency domain. The used data werernextracted from a subset of MIT-BIH arrhythmia data set, that contain 1000 ECG signals of 17rnclasses in total, collected from 45 patients. 12 classes were chosen from the subset whichrninclude Normal Sinus Rhythm (NSR), Atrial Premature Beat (APB), Atrial Flutter (AFL),rnAtrial Fibrillation (AFIB), Supraventricular Tachyarrhythmia (SVTA), Premature VentricularrnContraction (PVC), Ventricular Tachycardia (VT), Idioventricular Rhythm (IVR),rnVentricular Flutter (VFL), Left Bundle Branch Block Beat (LBBB), Right Bundle BranchrnBlock Beat (RBBB), and Pacemaker Rhythm (PR). The one dimensional ECG signals wererntransformed into joint time frequency spectrograms using Stockwell transform and intornscalograms using Continuous Wavelet Transform (CWT). By using different pretrainedrnnetworks for classifying spectrograms and scalograms namely GoogleNet, SqueezeNet, andrnResNet-50, different results were achieved. GoogleNet pretrained network showed the bestrnvrnperformance when using CWT generated scalograms with 93.85% accuracy, 96.42%rnprecision, 84.14% sensitivity, 99.36% average specificity and 89.86 F1-score. Based on thernresults, transfer learning especially GoogleNet proved to be efficient in classifying the twodimensionalrnscalograms of cardiac arrhythmias, while reducing the burden of trainingrnnetwork from scratch makes it easily applicable. Compared with recent techniques, resultsrnobtained using the proposed technique show the great promises of the 2-D CNN model inrnaccurate classification of arrhythmias using CWT and S-transform and the proposed methodrnresulted in higher accuracy and F1-score.