An image registration method for nanoimaging based on deep reinforcement learning

Abstract

The invention belongs to the technical field of nano-imaging, in particular to an image registration method for nano-imaging based on deep reinforcement learning. The method of the present invention includes constructing a deep reinforcement learning network model and using the network for image registration; the network model includes two branches; one branch includes a fully connected layer, and the input is an action sequence; one branch includes two convolutional layers and pooling Layer, the input is the selected reference picture and the picture to be registered; the output is the action probability distribution representing the strategy function; in the image registration part, 8 kinds of action sequences are designed to fine-tune the image to be registered; specifically include: The image is resampled; the image to be registered, the reference image and the resampled image are input into the constructed network model, and the probability distribution of the policy action is output. The invention has the advantages of fast speed, high precision, good robustness and strong adaptability; image registration is performed fully automatically, and the trouble of manual marking is eliminated.

Description

technical field [0001] The invention belongs to the technical field of nano-imaging, and in particular relates to an image registration method for nano-imaging. Background technique [0002] An Image Registration Method for Nanoscale Imaging Using Deep Reinforcement Learning Methods. Image registration is a method of using a certain algorithm to optimally map one or more pictures (locally) to a target picture based on a certain evaluation standard, which belongs to the field of software methods. [0003] The deep reinforcement learning method relates to the neural network structure in the deep learning and the control strategy of the reinforcement learning, and belongs to the field of software methods. [0004] Due to the jitter of the stepping motor and the influence of other factors such as X-ray exposure for a long time, each image formed after X-ray transmission will be slightly shifted at the crosshair. When processing images for nanoimaging, in order to eliminate the...

AI Technical Summary

Problems solved by technology

[0021] The traditional method will spend a lot of time on image feature detection and image feature matching, the running speed is slow, and the effect is often not good, which is greatly affected by the shooting effect of the sample

Although the manual marking method has high precision, it is difficult to prepare samples, heavy workload, low efficiency, high cost, and poor applicability

Images