Despite the promising achievements in fusion production, tokamaks still suffer from kinetic and MHD instabilities. In this work, we focus on Alfvén eigenmodes (AEs), a class of ubiquitous mixed kinetic and MHD instabilities, and present a data-driven pipeline to efficiently locate them in ECE spectrograms. ECE spectrograms have naturally high levels of noise, which must be reduced before further processing. For example, the classification and localizing rate of the AEs on raw spectrograms is about 60%. State-of-the-art deep learning techniques for enhancing the images need to see both noisy and clean versions of the data during the training which is something unavailable to us. To alleviate this issue, we first employ a pipeline of existing image processing techniques to partially denoise the spectrogram. These techniques, which include Gaussian filtering, median filtering, and morphological filtering, provide a baseline denoised image. The output of this pipeline is then used as a target for autoencoders to further enhance and segment the spectrograms using deep learning and autoencoder algorithms for localizing the AE instabilities.