Fig. 4

Comparison between the critical BTBT memory and reported optoelectronic memories in terms of proof-of-concept moving target tracking and recognition. a Reflective imaging using the customized “I”-shaped metal pattern, which can move in the 2D plane for image scanning. The laser pulse is 520 nm with a 200 μs duration. b The red arrow penetrates the same feature point across frames, representing its temporal movement. c–e Photocurrent mapping data scanned at different frames. p represents the same point of the customized “I”-shaped metal pattern, and the red dashed line contains feature point p and its neighbouring pixels. The feature point p, has the same movement trajectory as the whole, and target tracking is simplified to tracking the same points across frames, which is the basis of optical flow. f Double sets of moving trolley datasets obtained by the critical BTBT memory and other optoelectronic memories with an exposure time of 10 ms. The feature points in the two sets of time series images are detected by Harris corner detection and matched through the optical flow method. g The motion blur ratio is defined as the distance covered by the moving target during the exposure time divided by the length of the moving target. Relationship between the motion blur ratio and exposure time (also photomemory speed); different curves represent different trolley speeds. Motion blur exists when the exposure time is 10 ms, while almost no motion blur exists at 100 μs. Inset image: SEM view of the 4 × 4 critical BTBT memory and corresponding photomemory current in response to a 520 nm laser with different pulse widths, with red columns for 50 μs and blue for 100 μs. The solid lines are the results of fitting via a normal distribution. The preparation of arrays is shown in Supplementary Section 19. h The comparison of the accuracy between different neural networks, in which the training database is constructed by the critical BTBT memory and optoelectronic memories. Inset image: Loss function of the critical BTBT memory-based neural network and optoelectronic memory-based neural network