Learn, Denoise, and Discover: A guide to deep denoising with an application to electron microscopy

This tutorial provides a self-contained description of deep learning methodology for denoising, emphasizing aspects that are important in real-world scientific applications. The tutorial presents convolutional neural networks (CNNs) from first principles, explaining how to train them in a supervised and unsupervised manner. It also describes how to analyze the denoising strategies learned by these models and how to evaluate their performance. Illustrative examples are provided based on computational experiments with simple 1D piecewise constant signals, natural images, and simulated electron microscopy data. Code to reproduce all experiments is available online at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://kangningthu.github.io/AI-powered-denoising</uri>. In addition, a detailed case study is used to demonstrate the potential and challenges of applying deep denoising in practical scenarios. In the case study, supervised, unsupervised, and semisupervised deep learning models are leveraged to denoise transmission electron microscopy (TEM) data acquired at a very low signal-to-noise ratio (SNR) with the goal of investigating the dynamics of catalytic nanoparticles.