A terahertz compound eye based on a total diffraction neural network and a manufacturing method and application thereof

Abstract

The application relates to a terahertz compound eye based on a full-diffraction neural network and a manufacturing method and application, and relates to a terahertz compound eye and a manufacturing method and application. In order to solve the problems that a traditional compound eye architecture cannot detect target types and that the processing speed of a front-end sensing and computing module is limited, the compound eye comprises a plurality of full-optical-diffraction neural network channels, each channel comprises a plurality of unit structures, the unit structures are uniformly arranged, an external skeleton for fixing the diffraction channels and a detection array for receiving electric field signals at the rear end. The application innovatively constructs a terahertz compound eye based on a full-optical-diffraction neural network, recognizes the target types located at different directions of the compound eye, and simultaneously detects the distances between the targets and the compound eye. The terahertz compound eye based on the full-optical-diffraction neural network which is designed after sufficient physical network pre-training can effectively detect six different common targets in the surrounding environment and recognize three distance gradients. The application belongs to the technical field of optical sensing and computing.

Description

Technical Field

[0001] This invention relates to a terahertz compound eye, its manufacturing method, and its application, belonging to the field of optical sensing and computing technology. Background Technology

[0002] Compound eyes are biomimetic optical structures that mimic the visual system of insects. They are typically composed of multiple microlens units integrated in a specific array, featuring a large field of view, low distortion, and parallel information acquisition. They hold broad application prospects in target detection, environmental perception, intelligent navigation, and machine vision. Traditional compound eye systems mostly operate in the visible light band, relying primarily on optical structure design to acquire spatial information. However, limited by material properties and imaging methods, existing compound eye structures still have several shortcomings. On the one hand, traditional visible light compound eye systems have limited penetration capabilities through smoke, obstructions, and complex media environments, easily leading to decreased detection performance in low-light or complex environments. On the other hand, traditional compound eye systems can only perceive the target's orientation, making it difficult to detect the target type and its distance from the compound eye. As the number of compound eye channels increases, the system's data dimensionality significantly increases, and traditional signal processing methods gradually become insufficient to meet the application requirements of complex scenarios in terms of feature extraction efficiency and real-time processing capabilities. Terahertz waves, located between microwaves and infrared, possess strong penetrating power, low photon energy, and sensitivity to various biological tissues and materials. Introducing terahertz technology into compound eye systems not only expands the operating wavelength of the compound eye but also improves the system's target detection capabilities under conditions of obstruction, low visibility, and special media. Furthermore, all-optical diffraction neural networks have attracted widespread attention in recent years due to their ultra-high speed, low power consumption, and high parallel processing capabilities. They can rapidly analyze high-dimensional, multi-channel data, effectively improving the system's information processing speed.

[0003] Therefore, there is an urgent need to propose a terahertz compound eye design method and application based on all-optical diffraction neural networks to solve the above-mentioned technical problems. Summary of the Invention

[0004] To address the problems of traditional compound eye architectures being unable to detect target types and the limited processing speed of front-end perception and computing modules, this invention proposes a terahertz compound eye based on a fully diffractive neural network, along with its fabrication method and application.

[0005] The technical solution adopted by the present invention to solve the above problems is as follows: The terahertz compound eye based on the full diffraction neural network of the present invention includes a compound eye shell, and multiple diffraction sensing channels are provided inside the compound eye shell. Each diffraction sensing channel contains two physical hidden layers, and each physical hidden layer contains several dielectric unit structures. The several dielectric unit structures are evenly arranged. An electric field detection array is provided at the bottom of the compound eye shell. The electric field detection array is used to capture the output results of the full-optical diffraction neural network.

[0006] Furthermore, the all-optical diffraction neural network has 13 sensing channels, and each diffraction neural network has the same structural design, consisting of 30×30 phase modulation units with dielectric pillar structures.

[0007] Furthermore, several dielectric unit structures are arranged in a linear array, with each dielectric unit structure having the same lateral dimension of 1mm×1mm. By adjusting the height of the dielectric pillar unit, the phase of the incident light field can be controlled, thereby physically realizing an all-optical diffraction neural network.

[0008] Furthermore, the angle between two adjacent diffraction sensing channels is 25°, meaning that the field of view of terahertz in the horizontal and vertical directions is ±50 degrees.

[0009] The method for preparing a terahertz compound eye based on a fully diffractive neural network according to the present invention is achieved through the following steps: Step 1: Calculate the height of the dielectric pillar in the physical hidden layer of each diffraction channel by simulating and optimizing the phase distribution of the physical network based on the computing platform, and use a 3D printer to fabricate 13 sets of all-optical diffraction neural networks; Step 2: Use 3D printing technology to fabricate external slots for fixing and controlling the position of the all-optical diffraction neural network; Step 3: Integrated printing and processing of the compound eye shell.

[0010] The terahertz compound eye based on a full diffraction neural network described in this invention is applied to security inspection scenarios to achieve rapid real-time detection of dangerous targets.

[0011] Furthermore, the terahertz compound eye is also used in intelligent robot dogs to perform perception tasks.

[0012] The beneficial effects of this invention are: This invention constructs each compound eye channel based on an all-optical diffraction neural network, enabling the front end of the compound eye system to achieve ultra-high-speed, low-power information processing. This invention innovatively designs a terahertz compound eye based on an all-optical diffraction neural network, which can be flexibly mounted on smart cars and robot dogs to achieve rapid detection of several common dangerous items and is suitable for daily inspection tasks in public places. The terahertz compound eye based on all-optical diffraction neural network proposed in this invention has the advantages of high design flexibility, simultaneous detection of target type, orientation, and distance, and fast detection speed. Attached Figure Description

[0013] Figure 1 This is a side view of the overall structure of a terahertz compound eye based on an all-optical diffraction neural network. Figure 2 This is a physical image of a terahertz compound eye based on an all-optical diffraction neural network; Figure 3 This is a micrograph of the hidden layer in the terahertz compound eye channel based on an all-optical diffraction neural network; Figure 4 This is a schematic diagram of the hidden layer phase distribution training process of a terahertz compound eye based on an all-optical diffraction neural network; Figure 5 This is a schematic diagram of a test scenario for a smart car based on an application of a terahertz compound eye using an all-optical diffraction neural network. Figure 6 This is a schematic diagram of a test scenario mounted on a robot dog in an application based on a terahertz compound eye using an all-optical diffraction neural network. Figure 7 This is a schematic diagram of the reference channel output results (electric field intensity distribution) for different targets in an application based on an all-optical diffraction neural network terahertz compound eye. Figure 8 This is a schematic diagram of a display interface for environmental detection results of a terahertz compound eye based on an all-optical diffraction neural network. Detailed Implementation

[0014] Specific implementation method one: as follows Figures 1 to 4 As shown, the terahertz compound eye based on a full diffraction neural network described in this embodiment includes a compound eye shell, a plurality of diffraction sensing channels are provided inside the compound eye shell, each diffraction sensing channel contains two physical hidden layers, each physical hidden layer contains several dielectric unit structures, the several dielectric unit structures are evenly arranged, and an electric field detection array is provided at the bottom of the compound eye shell, the electric field detection array is used to capture the output results of the full-optical diffraction neural network.

[0015] The all-optical diffraction neural network has thirteen sensing channels, and each diffraction neural network has the same structural design, consisting of 30×30 phase control units with dielectric pillar structures.

[0016] In this structure, several dielectric unit structures are arranged in a linear array. Each dielectric unit structure has the same lateral dimension of 1mm×1mm. By adjusting the height of the dielectric column unit, the phase of the incident light field can be controlled, thereby physically realizing an all-optical diffraction neural network.

[0017] The angle between two adjacent diffraction sensing channels is 25°, meaning the field of view of terahertz in the horizontal and vertical directions is ±50 degrees.

[0018] Specific Implementation Method Two: The terahertz compound eye fabrication method based on a fully diffractive neural network described in this implementation method includes the following steps: Step 1: Calculate the height of the dielectric pillar in the physical hidden layer of each diffraction channel by simulating and optimizing the phase distribution of the physical network based on the computing platform. Use a 3D printer to fabricate 13 sets of all-optical diffraction neural networks. The step size of the 3D printer is 10 micrometers and the lateral processing error is 20 micrometers. Step 2: Use 3D printing technology to fabricate external slots for fixing and controlling the position of the all-optical diffraction neural network; Step 3: Integrated printing and processing of the compound eye shell.

[0019] Specific Implementation Method 3: This implementation method describes a terahertz compound eye based on a full diffraction neural network for use in security inspection scenarios, enabling rapid real-time detection of dangerous targets. The terahertz compound eye operates at a frequency of 0.1 THz. An external terahertz light source illuminates the target. When the target approaches the designed terahertz compound eye, the adjacent channel captures its corresponding incident light field, processes and calculates it, and the output is displayed as an electric field intensity distribution on the detection camera at the bottom. An externally connected personal laptop with an interactive interface allows for real-time monitoring of the terahertz compound eye's detection results.

[0020] In this embodiment, the terahertz compound eye detection result parsing method based on similarity matching is implemented through the following steps: Step 1: Obtain the electric field intensity distribution map of the output layer using a terahertz camera or electric field detection array; Step 2: Compare the obtained electric field intensity distribution map with the distribution map in the already stored labeled training set. Figure 1 A comparison is made and a comprehensive similarity score is calculated; Step 3: The detection result of the target with the highest similarity score is displayed on the interactive interface of the personal computer in the form of two-dimensional pixel blocks. The two-dimensional map contains the detection distance and the confidence information of the detection result.

[0021] The training set used to analyze the detection results contains the output layer electric field distribution corresponding to various targets at at least three detection distances. The front-end image acquisition rate is at least 30 frames per second, and the detection range of the back-end terahertz camera is at least ±4.8 cm. It can be applied to security inspection scenarios to achieve rapid real-time monitoring of at least three dangerous targets (such as pistols, daggers, axes, etc.). It can be mounted on a smart robot dog to perform perception tasks.

[0022] Figure 5 In this study, by mounting a multi-channel terahertz compound eye on a remotely controlled smart car, real-time monitoring of targets in the surrounding environment can be achieved. Figure 6In this study, by mounting a multi-channel terahertz compound eye on a remotely controlled smart car, it is possible to monitor dangerous targets hidden in various objects in the surrounding environment in real time. Figure 7 The test results of different hazardous targets in the baseline channel are shown in the output layer electric field distribution, with the focal points at different locations representing different targets. Figure 8 The final recognition results, including location, distance, and target type, are displayed through an interactive interface, obtained through the multi-channel distribution map on the left and the text output below, respectively. This invention constructs each compound eye channel based on an all-optical diffraction neural network, enabling ultra-high-speed, low-power information processing at the front end of the compound eye system. This invention innovatively designs a terahertz compound eye based on an all-optical diffraction neural network, which can be flexibly mounted on intelligent vehicles and robotic dogs to achieve rapid detection of several common dangerous items, suitable for daily inspection tasks in public places. The terahertz compound eye based on an all-optical diffraction neural network proposed in this invention has advantages such as high design flexibility, simultaneous detection of target type, orientation, and distance, and fast detection speed.

[0023] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent substitutions, and improvements made to the above embodiments without departing from the scope of the present invention, based on the technical essence of the present invention and within the spirit and principles of the present invention, shall still fall within the protection scope of the present invention.

Claims

1. A terahertz compound eye based on a fully diffractive neural network, characterized in that, It includes a compound eye shell, which contains multiple diffraction sensing channels. Each diffraction sensing channel contains two physical hidden layers. Each physical hidden layer contains several dielectric unit structures, which are evenly arranged. An electric field detection array is provided at the bottom of the compound eye shell. The electric field detection array is used to capture the output results of the all-optical diffraction neural network.

2. A terahertz compound eye based on a fully diffractive neural network according to claim 1, characterized in that, The all-optical diffraction neural network has thirteen sensing channels. Each diffraction neural network has the same structural design, consisting of 30×30 phase modulation units with dielectric pillar structures.

3. A terahertz compound eye based on a fully diffractive neural network according to claim 1, characterized in that, Several dielectric unit structures are arranged in a linear array. Each dielectric unit structure has the same lateral dimension of 1mm×1mm. By adjusting the height of the dielectric column unit, the phase of the incident light field can be controlled, thereby physically realizing an all-optical diffraction neural network.

4. A terahertz compound eye based on a fully diffractive neural network according to claim 1, characterized in that, The angle between two adjacent diffraction sensing channels is 0~100°, meaning that the field of view of terahertz in the horizontal and vertical directions is ±50 degrees.

5. A method for manufacturing a terahertz compound eye according to any one of claims 1 to 4, characterized in that, The specific steps include: Step 1: Calculate the height of the dielectric pillar in the physical hidden layer of each diffraction channel by simulating and optimizing the phase distribution of the physical network based on the computing platform, and use a 3D printer to fabricate 13 sets of all-optical diffraction neural networks; Step 2: Use 3D printing technology to fabricate external slots for fixing and controlling the position of the all-optical diffraction neural network; Step 3: Integrated printing and processing of the compound eye shell.

6. An application of the terahertz compound eye according to any one of claims 1 to 4, characterized in that, The terahertz compound eye is used in security inspection scenarios to achieve rapid, real-time detection of dangerous targets.

7. The application of a terahertz compound eye based on a fully diffractive neural network according to claim 6, characterized in that, The terahertz compound eye is also used in intelligent robot dogs to perform perception tasks.

Images