图1. 数据处理流程.(a)数值仿真生成的背景参考Z谱集；(b)采集得到的任一像素的原始Z谱，其中将红色点处的数据输入网络，以得到该像素的背景参考Z谱和水峰偏移量，用于后续的Z谱矫正；(c)任一仿真背景参考Z谱，其中红色点处的数据作为模型的训练输入，完整的Z谱（蓝线）以及水峰偏移量作为模型的训练输出；(d)使用的神经网络模型；(e)蓝紫色虚线为矫正后的背景参考Z谱，红色实线为矫正后的原始Z谱，箭头所指的阴影部分为两者相减得到的APT效应的贡献；(f)生成的APT图像

Fig.1. Schematic of data processing. (a) Simulated background Z-spectra; (b) The acquired raw Z-spectrum, in which the red solid dots are inputted into the neural network to produce background reference Z-spectrum and water peak offset for correcting Z-spectrum later; (c) A simulated background Z-spectrum, in which the data marked as red solid dots are inputs for training and the complete Z-spectrum represented by blue line as well as the water peak offset are targets for training; (d) The feedforward neural network generated and applied; (e) The water peak offset and the background Z-spectrum (marked as bluish violet dashed curve) are obtained from the network and used to correct the raw Z-spectrum. The shaded part indicated by the arrow is the difference between bluish violet dashed curve and red solid curve, which is contributed from APT effect; (f) The APT map