Fast AFM Imaging Based on Neural Network Compressed Sensing

Atomic force microscopy (AFM) realizes super-resolution imaging at micro-nano scale, but the time-consuming scanning imaging limits its measurement efficiency. Compression of AFM scanning points to improve the imaging speed has become a solution, and the imaging quality should also be guaranteed. In this work, the compressed sensing algorithm based on neural network is used as the reconstruction solver after compressed sampling of AFM data. The sample data of AFM database are incorporated into the network training set, which not only shortens the AFM scanning imaging time, but also ensures the high-resolution image reconstruction. Experimental results show that the network-based CS algorithm improves the peak signal noise ratio (PSNR) by 8.52 dB compared with the TVAL3 algorithm, and the minimum average reconstruction error rate (Err) is only 38.5 % of that of the TVAL3 algorithm. The reconstruction is fast, less than 0.5 s, and is less affected by the amount of sample data.