A Unified preprocessing framework for high-throughput diffraction pattern analysis

Four-dimensional scanning transmission electron microscopy (4D-STEM) is a high-throughput automated data acquisition technique with great potential for real-time data collection and analysis in automated STEM. However, its practical implementation is limited by challenges in data preprocessing, which hinder the timely and accurate interpretation of the large amounts of data it generates. Issues like pervasive noise, beam center drift, and elliptical distortions during high-throughput acquisition inevitably degrade diffraction patterns, leading to systematic errors in quantitative measurements. Conventional calibration algorithms are often material-specific and fail to provide a robust, generalizable solution. In this work, we introduce 4D-PreNet, an end-to-end deep-learning pipeline that integrates attention-enhanced U-Net and ResNet architectures to simultaneously perform denoising, center calibration, and ellipse calibration. The network is trained on extensive simulated datasets that cover a broad range of noise levels, drift magnitudes, and distortion types, thereby enabling generalization to experimental data obtained under different acquisition conditions. Quantitative evaluations demonstrate that 4D-PreNet reduces mean squared error by up to 50% in denoising and achieves sub-pixel center localization with average errors below 0.04 pixels. Compared to conventional algorithms, 4D-PreNet shows improved noise suppression and accurate restoration of diffraction features, enabling reliable real-time analysis of 4D-STEM data and supporting automated STEM workflows.