Smart jammer nodes can disrupt communication between a transmitter and a receiver in a wireless network, and they leave traces that are undetectable to classical jammer identification techniques, hidden in the time-frequency plane. These traces cannot be effectively identified through the use of the classical Fourier transform based time-frequency transformation (TFT) techniques with a fixed resolution. Inspired by the adaptive resolution property provided by the wavelet transforms, in this paper, we propose a jammer identification methodology that includes a pre-processing step to obtain a multi-resolution image, followed by the use of a classifier. Support vector machine (SVM) and deep convolutional neural network (DCNN) architectures are investigated as classifiers to automatically extract the features of the transformed signals and to classify them. Three different jamming attacks are considered, the barrage jamming that targets the complete transmission bandwidth, the synchronization signal jamming attack that targets synchronization signals and the reference signal jamming attack that targets the reference signals in an LTE downlink transmission scenario. The performance of the proposed approach is compared with the classical Fourier transform based TFT techniques, demonstrating the efficacy of the proposed approach in the presence of smart jammers.