CN-119832104-B - Asymmetric multifunctional imaging method and system for spatial diffraction depth neural network

CN119832104BCN 119832104 BCN119832104 BCN 119832104BCN-119832104-B

Abstract

The invention discloses a spatial diffraction depth neural network asymmetric multifunctional imaging method and system, which are suitable for the field of optical signal processing. The system consists of a signal input module, an all-optical diffraction neural network module and a signal acquisition module. The signal input module is responsible for providing a light source and a modulation element for generating and modulating the light field signal. The all-optical diffraction neural network module adopts a double-loss function optimization strategy to carry out asymmetric training. In particular, it introduces different loss functions in different transmission directions to train different functions simultaneously, thereby realizing diversified processing of signals. The signal acquisition module is responsible for acquiring and storing the final calculation result. The invention can realize different imaging functions according to different signal input directions, thereby adapting to the diversified demands of optical signal processing, filling the blank of the related field and providing a brand new thought for further research and application in the optical imaging field.

Inventors

WANG JIAN
MA XIAOXIAO

Assignees

华中科技大学

Dates

Publication Date: 20260508
Application Date: 20241220

Claims (8)

1. The asymmetric multifunctional imaging method of the spatial diffraction depth neural network is characterized by comprising the following steps of: The method comprises the steps of constructing a spatial diffraction neural network, wherein the spatial diffraction neural network is used as a carrier of optical imaging, so that an input light beam is imaged after being transmitted by the spatial diffraction neural network, the spatial diffraction neural network comprises n phase layers, and the phase combination of all the phase layers is (theta 1 ,θ 2 ……θ n ), wherein theta i is the phase of an ith phase layer, and i is more than or equal to 1 and less than or equal to n; The method comprises the steps of taking a first imaging output by an original input light beam after being input into the spatial diffraction neural network in a forward direction and a second imaging output by the original input light beam after being input into the spatial diffraction neural network in a backward direction as training data, training the spatial diffraction neural network by machine learning, determining phase combinations and obtaining a trained spatial diffraction neural network, wherein the training of the spatial diffraction neural network by the machine learning comprises the steps of adopting an asymmetric double-loss optimization strategy, wherein the double-loss optimization function of the asymmetric double-loss optimization strategy is as follows: Loss= αL 1 +β L 2 Where L 1 is the forward propagating loss function, L 2 is the backward propagating loss function, And Is a weighting coefficient; and inputting the input light beam to be imaged into the trained spatial diffraction neural network in a forward direction and a backward direction in sequence, and obtaining a first imaging and a second imaging.
2. The asymmetric multifunctional imaging system of the spatial diffraction depth neural network is characterized by comprising a signal input module, the spatial diffraction depth neural network and a signal output module; The signal input module is used for providing an input light beam carrying image information; The spatial diffraction depth neural network is used as a carrier of optical imaging, so that an input light beam is sequentially transmitted forward and backward through the spatial diffraction neural network and imaged to obtain a first imaging and a second imaging, the spatial diffraction neural network comprises n phase layers, the phase combination of all the phase layers is (theta 1 ,θ 2 ……θ n ), wherein theta i is the phase of an ith phase layer, i is more than or equal to 1 and less than or equal to n, the n phase layers are distributed according to a preset layer spacing, the phase combination (theta 1 ,θ 2 ……θ n ) on the phase layers is obtained by training by adopting a machine learning method, the layer spacing is not involved in optimization in the training process, the training of the spatial diffraction neural network by adopting the machine learning comprises the steps of adopting an asymmetric double-loss optimization strategy, and the double-loss optimization function of the asymmetric double-loss optimization strategy is as follows: Loss= αL 1 +β L 2 Where L 1 is the forward propagating loss function, L 2 is the backward propagating loss function, And Is a weighting coefficient; The signal output module is used for displaying imaging of the spatial diffraction depth neural network.
3. The spatial diffraction depth neural network asymmetric multi-functional imaging system of claim 2, wherein each phase layer comprises m x m pixels, each pixel for implementing from 0 to 2 Phase modulation in a range.
4. The spatial diffraction depth neural network asymmetric multi-functional imaging system of claim 2, wherein the processing material system of each phase layer comprises any one of gallium nitride, aluminum oxide, metal, silicon, photopolymer.
5. The spatial diffraction depth neural network asymmetric multi-functional imaging system of claim 2, wherein the medium between each phase layer is a vacuum, gas, liquid or solid.
6. The spatial diffraction depth neural network asymmetric multi-function imaging system of claim 2, wherein the input beam enters the spatial diffraction neural network after passing through a propagation medium, the propagation medium comprising air or a complex medium, the complex medium comprising a scattering medium, an absorbing medium, or a nonlinear medium.
7. The spatial diffraction depth neural network asymmetric multi-functional imaging system of claim 2, wherein the spatial diffraction depth neural network adapts light beams of different wavelengths, supporting asymmetric imaging operations of single-wavelength and multi-wavelength input light beams.
8. The asymmetric multifunctional imaging system of spatial diffraction depth neural network of claim 2, wherein the three-dimensional integration process for realizing the spatial diffraction depth neural network comprises electron beam lithography and maskless lithography.

Description

Asymmetric multifunctional imaging method and system for spatial diffraction depth neural network Technical Field The invention belongs to the field of optical signal processing, and particularly relates to a spatial diffraction depth neural network asymmetric multifunctional imaging method and system. Background The spatial diffraction deep neural network is a novel deep learning model which integrates the optical diffraction principle and the neural network algorithm. Optical diffraction is the phenomenon of light waves that are deflected and overlap when they encounter an obstacle or object edge. This phenomenon is an indispensable basis in optical imaging and information processing. By analyzing the diffraction pattern of the light waves, information about the shape, size, structure, etc. of the object can be obtained. The diffraction characteristic makes the diffraction have important application value in the fields of image processing, optical imaging, pattern recognition and the like. Second, neural networks are mathematical models that mimic the structure and function of human brain neurons. Through connection and weight adjustment among neurons, the neural network can learn and extract features from input data, so that functions of classification, identification, prediction and the like are realized. Deep learning is one of the fastest growing methods in the machine learning field, and is implemented in a computer using a multi-layer artificial neural network, and performs high-level functions by performing digital learning and abstraction on data, and even exhibits superior performance to human experts in some cases. In recent years, deep learning has made significant progress in the fields of medical image analysis, speech recognition, language translation, image classification, and the like. Combining optical diffraction and deep learning forms a spatial diffraction deep neural network. The combination can fully utilize the information processing capability of optical diffraction and the learning capability of a neural network, and realize faster processing and analysis of optical signals and images. However, the spatial diffraction depth neural network has high requirements on the environment, is easily influenced by factors such as light, temperature and humidity, and is limited in practical application. Therefore, how to explore the application capability of the spatial neural network and further expand the application scene of the spatial diffraction depth neural network is the key point of current research. Disclosure of Invention The invention provides a spatial diffraction depth neural network asymmetric multifunctional imaging method and a spatial diffraction depth neural network asymmetric multifunctional imaging system, which are characterized in that different functions can be flexibly executed according to different signal input directions, so that the imaging method and the imaging system are suitable for diversified requirements of optical signal processing. The design enables the imaging method to well cope with different functional requirements, and has wide application prospects. The innovation fills the research blank of the related technology in the direction of the spatial diffraction depth neural network, and brings new development opportunities to the fields of optical imaging and information processing. In order to achieve the aim, the invention provides a space diffraction depth neural network asymmetric multifunctional imaging method, which is used for constructing a space diffraction neural network, wherein the space diffraction neural network is used as a carrier of optical imaging, so that an input light beam is imaged after being transmitted through the space diffraction neural network, the space diffraction neural network comprises n phase layers, the phase combination of all the phase layers is (theta 1,θ2……θn), wherein theta i is the phase of an ith phase layer, and i is more than or equal to 1 and less than or equal to n; the method comprises the steps of taking a first imaging output by an original input light beam after being input into a space diffraction neural network in a forward direction and a second imaging output by the original input light beam after being input into the space diffraction neural network in a backward direction as training data, training the space diffraction neural network by machine learning, and determining a phase combination to obtain a trained space diffraction neural network; and inputting the input light beam to be imaged into the trained spatial diffraction neural network in a forward direction and a backward direction in sequence, and obtaining a first imaging and a second imaging. Preferably, training the spatial diffraction neural network using machine learning includes employing an asymmetric double-loss optimization strategy having a double-loss optimization function of: Loss=αL1+βL2 Where L 1 is the forward propagation lo