CN-122023820-A - Atomic-level precision surface feature recognition and positioning method based on deep learning multi-mode image fusion

CN122023820ACN 122023820 ACN122023820 ACN 122023820ACN-122023820-A

Abstract

The application relates to the technical field of intelligent nano image analysis. The application discloses an atomic-level precision surface feature recognition and positioning method based on deep learning multi-mode image fusion, which can improve recognition efficiency and accuracy of various adsorption molecules and defects on the surface of a crystal and realize automatic recognition and accurate positioning of the various adsorption molecules and the defects on the surface of the crystal. The method comprises the steps of obtaining a target fusion image, wherein the target fusion image is formed by fusing a positive bias scanning image and a negative bias scanning image, the positive bias scanning image is an image of the surface of a crystal collected by a scanning tunnel microscope in a positive bias mode, the negative bias scanning image is an image of the surface of the crystal collected by the scanning tunnel microscope in a negative bias mode, and inputting the target fusion image into an atomic-level precision surface feature identification and positioning model for positioning detection treatment to obtain identification and positioning results of various adsorbed molecules and defects on the surface of the crystal.

Inventors

CUI YUWEI
WU KUNRONG
YANG LUYAN
Duan Mingchao
WANG GUANYONG
Pan Tianluo
WU PING
YAO JUAN
HE YU

Assignees

深圳国际量子研究院
深圳大学

Dates

Publication Date: 20260512
Application Date: 20251226

Claims (10)

1. An atomic-level precision surface feature recognition and positioning method based on deep learning multi-mode image fusion is characterized by comprising the following steps: acquiring a target fusion image, wherein the target fusion image is formed by fusing a positive bias scanning image and a negative bias scanning image, the positive bias scanning image is an image of the surface of a crystal acquired by a scanning tunnel microscope in a positive bias mode, and the negative bias scanning image is an image of the surface of the crystal acquired by the scanning tunnel microscope in a negative bias mode; and inputting the target fusion image into an atomic-level precision surface feature recognition and positioning model for recognition and positioning detection treatment to obtain recognition and positioning results of various adsorption molecules and defects on the surface of the crystal.
2. The method for identifying and locating surface features with atomic level precision based on deep learning multimodal image fusion as defined in claim 1, wherein the step of obtaining the target fusion image comprises: acquiring the positive bias scan image and the negative bias scan image; and adopting a multi-mode fusion network to fuse the positive bias scanning image and the negative bias scanning image to obtain the target fusion image.
3. The atomic-level-precision surface feature recognition and localization method based on deep learning multi-modal image fusion according to claim 2, wherein the step of performing fusion processing on the positive bias scan image and the negative bias scan image by using a multi-modal fusion network to obtain the target fusion image comprises: adopting a first image feature extraction module of the multi-mode fusion network to carry out image feature extraction processing on the forward bias scanning image to obtain a first basic feature image and a first detail feature image; Performing image feature extraction processing on the negative bias scanning image by adopting the first image feature extraction module to obtain a second basic feature image and a second detail feature image; And carrying out fusion and decoding processing on the first basic feature image, the first detail feature image, the second basic feature image and the second detail feature image by adopting an image fusion module of the multi-mode fusion network to obtain the target fusion image.
4. The atomic level precision surface feature recognition and positioning method based on deep learning multi-modal image fusion according to claim 3, wherein the step of obtaining a first basic feature image and a first detail feature image by performing image feature extraction processing on the forward bias scan image by using a first image feature extraction module of the multi-modal fusion network comprises: adopting a first shared encoder of the first image feature extraction module to extract image features of the forward bias scanning image to obtain a first feature image to be processed; adopting a first basic encoder of the first image feature extraction module to extract image features of the first feature image to be processed, and obtaining the first basic feature image; And carrying out image feature extraction processing on the first feature image to be processed by adopting a first detail encoder of the first image feature extraction module to obtain the first detail feature image.
5. The atomic level precision surface feature recognition and positioning method based on deep learning multi-modal image fusion according to claim 3, wherein the step of using the image fusion module of the multi-modal fusion network to fuse and decode the first basic feature image, the first detail feature image, the second basic feature image, and the second detail feature image, and obtaining the target fusion image comprises: A second basic encoder of the image fusion module is adopted to carry out fusion processing on the first basic feature image and the second basic feature image, so as to obtain a basic fusion image; Adopting a second detail encoder of the image fusion module to fuse the first detail characteristic image and the second detail characteristic image to obtain a detail fusion image; and adopting a decoder of the image fusion module to decode the basic fusion image and the detail fusion image to obtain the target fusion image.
6. The atomic-level precision surface feature recognition and positioning method based on deep learning multi-modal image fusion according to claim 1, wherein the step of inputting the target fusion image into an atomic-level precision surface feature recognition and positioning model for recognition and positioning detection processing to obtain recognition and positioning results of a plurality of adsorption molecules and defects on the surface of the crystal comprises the following steps: inputting the target fusion image into a backbone network of the atomic-level precision surface feature recognition and positioning model for multi-scale analysis processing to obtain a plurality of images to be processed with different scales; Inputting the images to be processed with different scales into a neck module of the atomic-level precision surface feature recognition and positioning model to perform aggregation treatment to obtain a target aggregation image; And inputting the target aggregate image into a detection head module of the atomic-level precision surface feature recognition and positioning model to perform positioning detection processing, so as to obtain recognition and positioning results of various adsorption molecules and defects on the surface of the crystal.
7. The method for recognizing and locating surface features with atomic-level precision based on deep learning multi-modal image fusion according to claim 1, wherein the plurality of adsorption molecules are PH molecules, PH 2 molecules and P atoms, and the defect is a point defect, a dimer defect or a multi-dimer defect.
8. An atomic-level precision surface feature recognition and positioning device based on deep learning multi-mode image fusion is characterized by comprising the following components: the target fusion image acquisition module is used for acquiring a target fusion image, wherein the target fusion image is formed by fusing a positive bias scanning image and a negative bias scanning image, the positive bias scanning image is an image of the surface of a crystal acquired by a scanning tunnel microscope in a positive bias mode, and the negative bias scanning image is an image of the surface of the crystal acquired by the scanning tunnel microscope in a negative bias mode; The recognition and positioning detection module is used for inputting the target fusion image into an atomic-level precision surface feature recognition and positioning model to perform recognition and positioning detection processing, so as to obtain recognition and positioning results of various adsorption molecules and defects on the surface of the crystal.
9. A terminal device comprising a processor and a memory for storing a computer program, said processor for invoking and running the computer program stored in said memory, performing the steps of the deep learning multi-modal image fusion based atomic level precision surface feature identification and localization method according to any one of the preceding claims 1 to 7.
10. A computer-readable storage medium storing a computer program for causing a computer to perform the steps of the atomic level precision surface feature recognition and localization method based on deep learning multimodal image fusion as claimed in any one of the preceding claims 1 to 7.

Description

Atomic-level precision surface feature recognition and positioning method based on deep learning multi-mode image fusion Technical Field The application relates to the technical field of intelligent nano image analysis. More particularly, the application relates to an atomic-level precision surface feature recognition and positioning method based on deep learning multi-mode image fusion. Background The traditional atomic-level precision surface feature recognition and positioning method is characterized in that various adsorption molecules and defects on the surface of a crystal are recognized by manually interpreting a Scanning Tunneling Microscope (STM) image, and the method has the bottlenecks of low efficiency, strong subjectivity, difficulty in realizing automatic recognition and positioning of the various adsorption molecules and the defects, and the like, so that the prior art needs to be improved and improved. Disclosure of Invention The embodiment of the application aims to provide an atomic-level precision surface feature recognition and positioning method based on deep learning multi-mode image fusion, which can improve recognition efficiency and accuracy of various adsorption molecules and defects on the surface of a crystal and realize automatic recognition and accurate positioning of the various adsorption molecules and the defects on the surface of the crystal. The embodiment of the application is mainly realized by the following technical scheme: In a first aspect of the embodiment of the present application, there is provided an atomic-level precision surface feature recognition and positioning method based on deep learning multi-modal image fusion, including: acquiring a target fusion image, wherein the target fusion image is formed by fusing a positive bias scanning image and a negative bias scanning image, the positive bias scanning image is an image of the surface of a crystal acquired by a scanning tunnel microscope in a positive bias mode, and the negative bias scanning image is an image of the surface of the crystal acquired by the scanning tunnel microscope in a negative bias mode; and inputting the target fusion image into an atomic-level precision surface feature recognition and positioning model for recognition and positioning detection treatment to obtain recognition and positioning results of various adsorption molecules and defects on the surface of the crystal. According to one embodiment of the present application, the step of acquiring the target fusion image includes: acquiring the positive bias scan image and the negative bias scan image; and adopting a multi-mode fusion network to fuse the positive bias scanning image and the negative bias scanning image to obtain the target fusion image. According to one embodiment of the present application, the step of fusing the positive bias scan image and the negative bias scan image using a multi-modal fusion network to obtain the target fusion image includes: adopting a first image feature extraction module of the multi-mode fusion network to carry out image feature extraction processing on the forward bias scanning image to obtain a first basic feature image and a first detail feature image; Performing image feature extraction processing on the negative bias scanning image by adopting the first image feature extraction module to obtain a second basic feature image and a second detail feature image; And carrying out fusion and decoding processing on the first basic feature image, the first detail feature image, the second basic feature image and the second detail feature image by adopting an image fusion module of the multi-mode fusion network to obtain the target fusion image. According to one embodiment of the present application, the step of performing image feature extraction processing on the forward bias scan image by using the first image feature extraction module of the multi-modal fusion network to obtain a first basic feature image and a first detail feature image includes: adopting a first shared encoder of the first image feature extraction module to extract image features of the forward bias scanning image to obtain a first feature image to be processed; adopting a first basic encoder of the first image feature extraction module to extract image features of the first feature image to be processed, and obtaining the first basic feature image; And carrying out image feature extraction processing on the first feature image to be processed by adopting a first detail encoder of the first image feature extraction module to obtain the first detail feature image. According to one embodiment of the present application, the step of performing fusion and decoding processing on the first basic feature image, the first detail feature image, the second basic feature image, and the second detail feature image by using the image fusion module of the multimodal fusion network, and obtaining the target fusion image includes: A second