CN-122028527-A - Dual-mode polarization fusion visual sensor, preparation method thereof and model construction method

CN122028527ACN 122028527 ACN122028527 ACN 122028527ACN-122028527-A

Abstract

The invention relates to the technical field of photoelectricity, in particular to a bimodal polarization fusion vision sensor, a preparation method thereof and a model construction method. Meanwhile, the van der Waals heterojunction formed by the ferroelectric layer and the anisotropic photosensitive layer simulates the plasticity behavior of biological synapses through charge recombination-capturing dynamics under polarized light irradiation, and can asynchronously output polarized dynamic images with high time resolution and low delay in an event-driven mode. The visual sensor integrates static imaging and polarization dynamic perception in a single device at a hardware level, remarkably improves visual perception robustness and information richness in a complex light field (such as glare and haze), and provides a bionic visual solution with high energy efficiency and high performance for automatic driving, machine vision and other applications.

Inventors

LIANG QIJIE
Ning Liqiong
LI WENBO
ZENG QUN

Assignees

松山湖材料实验室
华南师范大学

Dates

Publication Date: 20260512
Application Date: 20260123

Claims (10)

1. The dual-mode polarization fusion vision sensor is characterized by comprising a bottom electrode layer, a ferroelectric layer, an anisotropic photosensitive layer and a top electrode layer which are sequentially stacked, wherein the ferroelectric layer forms a vertical transistor and a van der Waals heterojunction with the photosensitive layer; the ferroelectric layer has a bulk photovoltaic effect and ferroelectricity and is used for static imaging under light irradiation; The van der Waals heterojunction is used for simulating biological synapse behavior under polarized light irradiation through charge recombination and capturing dynamics, and polarization dynamic imaging is performed based on an event-driven principle.
2. The bimodal polarization fusion vision sensor of claim 1, wherein the ferroelectric layer comprises any one of a 3R-MoS 2 ferroelectric layer, an InSe ferroelectric layer, an In 2 Se 3 ferroelectric layer, a ReS 2 ferroelectric layer, and a ReSe 2 ferroelectric layer.
3. The bimodal polarization fusion vision sensor of claim 1, wherein the photosensitive layer comprises any one of a PdSe 2 photosensitive layer, a BP photosensitive layer, a B-AsP photosensitive layer, a SnS photosensitive layer, a SnSe photosensitive layer, a GeS photosensitive layer, and a GeSe photosensitive layer.
4. The dual-mode polarization fusion vision sensor of claim 1, wherein the static imaging is in a self-powered mode and power consumption is zero, and the power consumption of the polarization dynamic imaging is less than a preset value.
5. The bimodal polarization fusion vision sensor of claim 1, wherein the conductance of the vision sensor changes when the light pulse of polarized light changes to achieve synaptic weight control of synaptic behavior.
6. The bimodal polarization fusion vision sensor of claim 1, wherein the simulated biological synaptic behavior comprises short-term plasticity, dipulse facilitation, long-term plasticity, and electrical pulse suppression behavior.
7. A method for constructing a motion direction recognition network model, the method comprising: acquiring an imaging image acquired when the bimodal polarization fusion vision sensor as claimed in any one of claims 1 to 6 is used for moving a moving object; constructing a training set by adopting the imaging image and the motion direction of the corresponding moving object; training a preset neural network by adopting the training set to obtain a motion direction identification network model.
8. The method of claim 7, wherein the recognition accuracy of the motion direction recognition model is greater than a threshold when the polarized light is a preset polarization direction.
9. A method for preparing a bimodal polarization fusion vision sensor, the method comprising: stripping the materials for preparing the ferroelectric layer and the anisotropic photosensitive layer by adopting a mechanical stripping method to obtain a two-dimensional material sample, wherein the two-dimensional material sample comprises a ferroelectric layer sample and a photosensitive layer sample which are respectively formed on PDMS; Sequentially transferring the ferroelectric layer sample and the photosensitive layer sample to a substrate comprising a bottom electrode layer, wherein a vertical transistor is formed on the bottom electrode layer by the ferroelectric layer, and the ferroelectric layer and the photosensitive layer form a van der Waals heterojunction; A top electrode layer is formed on a surface of the vertical transistor and van der waals heterojunction remote from the bottom electrode layer.
10. The method of manufacturing according to claim 9, wherein the bottom electrode layer comprises a chromium layer and a gold layer which are stacked, and the top electrode layer comprises a source electrode and a drain electrode.

Description

Dual-mode polarization fusion visual sensor, preparation method thereof and model construction method Technical Field The invention relates to the technical field of photoelectricity, in particular to a bimodal polarization fusion vision sensor, a preparation method thereof and a model construction method. Background With the rapid development of the fields of computer vision, robot perception, intelligent systems and the like, increasingly stringent requirements are put on the precision, robustness and timeliness of environment perception, and effective hardware is required to detect static gray scale, dynamic events and polarization information in a visual scene. Machine vision based on frame capture, such as active pixel sensors (Active Pixel Sensor, APS), is capable of static imaging, but has inherent limitations in processing high-speed dynamic scenes, i.e., difficulty in capturing transient changes or polarization characteristics of moving objects. Meanwhile, the dynamic vision sensor (Dynamic Vision Sensor, DVS) is used as a bionic vision device, can asynchronously output pixel-level brightness change events in a manner of microsecond time resolution, high dynamic range and low power consumption, and has unique advantages in terms of high-speed motion perception and low delay response. However, DVS responds only to luminance changes, lacks complete characterization capabilities for static scenes, and loses key optical information such as color and polarization. Dynamic vision and active pixel sensors (DYNAMIC AND ACTIVE-pixel Vision Sensor, DAVIS) enable static and dynamic bimodal imaging. The CMOS integrated organic neuromorphic imager for high-resolution bimodal imaging in the prior art realizes the switching between standard mode static imaging and synaptic mode dynamic imaging by adjusting the weight ratio of the imager receptor. And two photocurrent modes are generated by using the thickness difference of the h-BN double gate through the charge coupling effect, so that synchronous detection of dynamic and static images can be realized in the phototriode. Nevertheless, the development of DAVIS capable of converged polarization detection remains limited, on the one hand, static and dynamic mode switching is faced with high energy consumption, data redundancy and clock synchronization problems, and on the other hand, its perceptibility is often significantly limited in complex lighting (glare), low contrast or scenes where scattering media (such as fog, smoke, underwater, etc.) are present. Therefore, there is an urgent need to develop a bimodal visual sensor capable of fusing polarization detection. Disclosure of Invention In view of the above, the invention provides a bimodal polarization fusion vision sensor, a preparation method thereof and a model construction method thereof, so as to solve the technical problem that the bimodal vision sensor with synchronous fusion polarization detection and clock is lacking in the prior art. The technical scheme provided by the invention is as follows: The invention provides a bimodal polarization fusion vision sensor, which comprises a bottom electrode layer, a ferroelectric layer, an anisotropic photosensitive layer and a top electrode layer which are sequentially stacked, wherein the ferroelectric layer forms a vertical transistor and forms a van der Waals heterojunction with the photosensitive layer, the ferroelectric layer has a bulk photovoltaic effect and ferroelectricity and is used for static imaging under irradiation of polarized light, and the van der Waals heterojunction is used for simulating biological synaptic behavior through charge recombination and capturing dynamics under irradiation of polarized light and is used for polarized dynamic imaging based on an event-driven principle. Meanwhile, the van der Waals heterojunction formed by the ferroelectric layer and the anisotropic photosensitive layer simulates the plasticity behavior of biological synapses through charge recombination-capturing dynamics under polarized light irradiation, and can asynchronously output polarized dynamic images with high time resolution and low delay in an event-driven mode. The visual sensor integrates static imaging and polarization dynamic perception in a single device at a hardware level, remarkably improves visual perception robustness and information richness in a complex light field (such as glare and haze), and provides a bionic visual solution with high energy efficiency and high performance for automatic driving, machine vision and other applications. In an alternative embodiment, the ferroelectric layer comprises any one of a 3R-MoS 2 ferroelectric layer, an InSe ferroelectric layer, an In 2Se3 ferroelectric layer, a ReS 2 ferroelectric layer, and a ReSe 2 ferroelectric layer. According to the invention, the ferroelectric performance and the bulk photovoltaic effect of the device can be customized and optimized according to different application scenes