A high-ratio lossy data compression framework for beam diagnostic systems using ResNet AutoEncoder networks

Modern particle accelerator facilities employ sophisticated beam diagnostic systems (BDS) for operation status monitoring, in which advanced beam detectors generate high-fidelity data streams with a throughput exceeding hundreds of MB/s, posing substantial storage and transmission challenges. To address the bottleneck, this article presents a novel lossy data compression method based on ResNet AutoEncoder (RAE) architecture that achieves high compression ratios for beam diagnostic data. By combining the dimensionality reduction capabilities of AutoEncoders with the feature extraction power of ResNet, the proposed framework achieves a remarkable baseline compression ratio of 256 × , surpassing current state-of-the-art (SOTA) methods with reconstruction quality of SSIM: 78.67 ± 9.54 % and PSNR: 24.98 ± 1.68 dB. The integration of least absolute shrinkage and selection operator (LASSO) regularization during the training process enables additional sparse coding optimization, further enhancing compression performance, achieving an exceptional 315 × compression ratio while maintaining reconstruction quality with SSIM: 73.74 ± 8.90 %, PSNR: 24.55 ± 1.63 dB, and a Beam Statistical Relative Error (BSRE) of 5.82 ± 4.83 % for critical beam parameters. The practical feasibility of the proposed method is validated on measured beam data from HUST-PTF, potentially reducing the transmission bandwidth of scintillation screen data from 300 MB/s to below 1 MB/s. The proposed RAE compression framework offers significant improvements over state-of-the-art methods, enabling substantial infrastructure cost reductions for particle accelerator facilities.