Modern industrial drives use induction motors (IM) due to their starting and running torque requirements and four-quadrant operation. The voltages and currents of each of the three phases as well as the acceleration and velocity signals may be used to identify failures in the rotor bars of the motor. In the past, traditional signal processing-based feature extraction approaches and machine learning algorithms have been used for the diagnosis of the number of broken rotor bars for a failed IM. In this paper, a new scheme is investigated for the determination of the number of broken rotor bars. Specifically, the deep learning approaches are examined for the mentioned purposes. To this end, transfer learning approaches on the convolutional neural network (CNN) are presented. For denoising, initially, a bandpass filter is employed, and then the signals are converted to time-frequency images by using the continuous wavelet transform (CWT). The obtained images are used for the fine-tuning of the pre-trained ResNet18 model and the deep feature extraction and classification with the support vector machine (SVM) classifier. The classification accuracy is used for performance evaluation metrics. Moreover, the experiments show that the highest accuracy score of 100% is obtained with the deep features that are extracted from the mechanical vibration signal and current signal. On the other hand, a performance comparison with the published approaches is also carried out. Finally, the comparisons show that the proposed method produces better accuracy scores than the compared methods.