Automatic identification of proton whistler crossing frequency and its application in ZH-1 satellite

The cross-frequency of proton whistler (PW) is a critical parameter for inverting the spatial distribution of hydrogen ions in space. Currently, this parameter is mainly extracted manually. To address the challenges posed by the vast satellite data, we propose the Proton Whistler cross-frequency Network (PWNet), a deep learning-based framework for the automatic extraction of proton whistler cross-frequency. The main steps are as follows: (1) Apply a 4-second sliding window to perform short-time Fourier transform on the waveform data from the Electric Field Detector (EFD) onboard the Zhangheng-1 satellite into time-frequency spectrograms; (2) Construct a multi-task deep learning model that integrates lightweight convolutional structures and pruning techniques; (3) Input time-frequency images into the model to achieve automatic identification of proton whistler events and their cross-frequencies. The model's performance in detecting cross-frequency was evaluated using EFD measurement data from January 6—10, 2020, demonstrating an average precision of 98.0%, with accuracy and recall reaching 95.4% and 93.8%, respectively. The model features only 1.01 × 106 parameters and requires 2.9 GFLOPS computational power, achieving "compact model" level lightweight metrics. Applied PWNet to full-orbit data of February 2020, PWNet reveals a spatiotemporal distribution characteristics: Cross-frequency values increase with geomagnetic latitude, and PW occurrence shows higher density in the Northern Hemisphere (40°—60° geomagnetic latitude band) than in the corresponding Southern Hemisphere region, consistent with the theoretical prediction of hemispheric asymmetry in PW distributions. This validates the effectiveness of the model, providing a reliable intelligent solution for large-scale analysis of proton whistler cross-frequency events, and holds significant reference value for exploring other space electromagnetic wave phenomena.