Angstrom resolution low-altitude spectrum remote sensing chip, system and device

Abstract

The invention provides a sub-angstrom resolution low-altitude spectrum remote sensing chip, system and device. The problem that sub-angstrom molecular features cannot be recognized due to the fact that fine spectral differences are lost in a complex scene due to insufficient resolution (larger than or equal to 0.3 nm) of existing spectral remote sensing equipment can be solved; a fixed compression strategy cannot adapt to the spectral complexity of ground objects, important information loss or invalid data redundancy; the problems that mechanical scanning or data processing delay restricts imaging timeliness, laboratory off-line interpretation delay is large, agricultural / disaster monitoring real-time requirements cannot be met, and the size and the weight are large and are difficult to adapt to an unmanned aerial vehicle platform are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of spectral remote sensing systems, in particular to an amplitude resolution low-altitude spectral remote sensing chip, system and device. BACKGROUND

[0002] As a core means of remote sensing detection, spectral remote sensing technology is widely used in environmental monitoring, precision agriculture, mineral exploration and other fields. In related technologies, scanning spectrometers based on grating dispersion principle and Fourier transform spectrometers (FTS) constitute the current mainstream technology system. Specifically, the grating dispersion type system realizes wavelength separation by the cooperative operation of the incident slit, collimating mirror, diffraction grating and CCD detector through the grating equation; and the FTS system relies on the multi-path transmission characteristics of the Michelson interferometer to generate an interference pattern through dynamic mirror displacement and analyze the spectrum through Fourier transform. With the popularization of unmanned aerial vehicle platforms, low-altitude remote sensing technology has gone from the laboratory to practical application, and its technical development has gone through the evolution from mechanical scanning to static interference, covering the complete technology chain from optical design to data processing, including key links such as spectral acquisition, spatial imaging, signal analysis, etc.

[0003] However, the existing spectral remote sensing system has significant limitations in technical implementation. Specifically, the grating dispersion type device is limited by the contradiction between the slit light efficiency and the resolution. When a high-density grating with a density of 2400 lp / mm or more is used to achieve a resolution of 0.5 nm, the grating size needs to reach 150 mm, resulting in a sharp increase in system volume and weight. Although the FTS system has high throughput advantage, its interference module has strict requirements on vibration tolerance (±0.1 μm), and under the working condition of unmanned aerial vehicle platform measurement vibration offset ≥1 μm, the spectral analysis accuracy decreases significantly. Based on this, the existing technology faces fundamental challenges in dynamic scene monitoring: The disadvantages of the grating type include: (1) low light efficiency (low flux): this is its most famous disadvantage - slit limitation. In order to obtain high resolution, very narrow entrance slits must be used, which will block most of the light, resulting in a sharp drop in light flux. This is very unfavorable for the detection of weak light signals (such as fluorescence, astronomical spectra). (2) Slow scanning speed: for scanning grating spectrometer (by rotating the grating to measure each wavelength point sequentially), it takes a long time to obtain a whole spectrum, which cannot be used for the study of fast-changing transient processes. (3) Stray light problem: defects of the grating and non-ideal reflection / diffraction of optical devices can produce stray light, i.e. light of non-measured wavelengths is received by the detector. This can cause the baseline of the spectrum to be raised and distorted, especially near strong spectral lines or when measuring weak absorption peaks, affecting the measurement accuracy. (4) Resolution and flux conflict: resolution and light flux are a pair of difficult to reconcile. Improving the resolution (narrowing the slit) will inevitably reduce the flux, and vice versa. The disadvantages of the Fourier type include: (1) complex structure, high cost: high-precision dynamic mirror scanning mechanism, stable interferometer and powerful computing unit are needed for real-time Fourier transform, and the manufacturing and maintenance cost is usually higher than that of the same level of grating spectrometer. (2) Extremely high requirement for mechanical stability: the scanning of the dynamic mirror must maintain excellent linearity and stability, any slight vibration or tilt will cause the interference pattern to be distorted, thereby affecting the quality of the final spectrum. (3) There is a dynamic range limit: the detector needs to respond linearly to the extremely strong interference signal at zero optical path difference and the extremely weak signal at a long distance within one scanning period, which puts high requirements on the linear dynamic range of the detector. SUMMARY

[0004] The present application aims to at least partially solve one of the technical problems in the related art.

[0005] To this end, the first object of the present application is to provide an airmass resolution sub-Airy unit low-altitude spectral remote sensing system.

[0006] The second object of the present application is to provide an integrated pod device based on the airmass resolution sub-Airy unit low-altitude spectral remote sensing system.

[0007] The third object of the present application is to provide an encoding modulation chip for airmass resolution sub-Airy unit spectral image snapshot acquisition.

[0008] The fourth object of the present application is to provide an airmass resolution sub-Airy unit synchronous spectral imaging method.

[0009] To achieve the above objects, the first aspect of the present application provides an airmass resolution sub-Airy unit low-altitude spectral remote sensing system, comprising an optical imaging module, a spectral acquisition module, a storage communication module and a comprehensive control module; wherein, The optical imaging module is used for capturing spatial image information of a target area and pre-processing incident light rays to provide high-quality incident light to the spectrum acquisition module, and simultaneously transmitting the split spectrum or independently collected real-time image to a rear system to provide visual guidance for remote control of the remote sensing platform. The spectrum acquisition module is used for decomposing the incident light from the optical imaging module into a spectrum with sub-angstrom resolution and converting the spectrum into a corresponding electrical signal. The storage communication module is used for storing spectrum data represented by the electrical signal output by the spectrum acquisition module in real time, and performing bidirectional data interaction with the remote sensing platform, and the interaction content includes flight parameters, system pose information, ground control instructions and data link state information. The integrated control module is used for coordinating the cooperative work of the optical imaging module, the spectrum acquisition module and the storage communication module, and performing environmental adaptive compensation control, real-time data processing and servo control; wherein the environmental adaptive compensation control includes self-stabilization control, the real-time data processing includes encoding of images and information, and the servo control is used for responding to ground instructions and driving the remote sensing platform to adjust the attitude.

[0010] To achieve the above purpose, the second aspect of the present application provides an integrated pod device based on a sub-angstrom resolution low-altitude spectral remote sensing system, at least including a central axis bundle, a base, an optical cabin and an integrated control board, wherein, The central axis bundle extends longitudinally, and its upper end is connected to the unmanned aerial vehicle platform through the base, and its lower end is connected to the optical cabin, the base is integrated with an azimuth angle sensor and an azimuth motor, the azimuth angle sensor is used for collecting azimuth angle information of the base relative to the unmanned aerial vehicle platform in real time and transmitting the information to the integrated control board, and the azimuth motor is used for receiving driving instructions from the integrated control board to drive the entire pod to rotate around the central axis bundle to adjust the overall orientation of the pod. The optical cabin is located at the lower part of the pod, and is integrated with a sub-angstrom spectrum acquisition device, a visible light camera and an industrial computer inside; the sub-angstrom spectrum acquisition device is equipped with a sub-angstrom resolution low-altitude spectral remote sensing chip and is used for collecting high-spectral resolution spectrum data of the target area; the optical axis of the visible light camera is parallel or coaxial with the viewing axis of the sub-angstrom spectrum acquisition device, and is used for synchronously collecting real-time visible light images of the target area to provide image transmission visual guidance for remote control of the unmanned aerial vehicle; the industrial computer is in communication connection with the unmanned aerial vehicle platform and the integrated control board, and is used for receiving control instructions, coordinating and controlling the working states of the sub-angstrom spectrum acquisition device and the visible light camera, and storing and forwarding the collected spectrum data and visible light images in real time. The integrated control board is integrated in the optical cabin, communicates with the flight control system and the image transmission system of the unmanned aerial vehicle platform through an open interface, receives flight parameters, control instructions and time synchronization signals from the unmanned aerial vehicle, and also connects and processes feedback signals of a pitch angle sensor, a pitch speed sensor and an azimuth speed sensor arranged in the optical cabin, and calculates complete position information of the optical cabin in real time; based on the position information and the received control instructions, the integrated control board executes a servo control algorithm, generates a driving signal and outputs the driving signal to an azimuth motor and a pitch motor in the optical cabin, so that stable pointing and dynamic tracking of the optical cabin in two degrees of freedom of azimuth and pitch are realized, and stable imaging and remote sensing detection of a target area are completed.

[0011] To achieve the above purpose, the third aspect of the present application provides an encoding modulation chip for sub-Angstrom spectral image snapshot acquisition, comprising a silicon substrate, an array of tunable light filtering units integrated on the silicon substrate, and an integrated AI inference processing unit; The array of tunable light filtering units is composed of a silicon nitride waveguide layer and an array of metal resonant cavities stacked on the silicon substrate, each metal resonant cavity serving as an independent light filtering unit, realizing a non-correlation designed spectral transmittance response by adjusting its physical structure, physically encoding and modulating the incident light, so as to break the spectral correlation between pixels in traditional imaging, thereby synchronously capturing multi-dimensional spectral information with a resolution of 0.1 nm under single exposure; The silicon nitride waveguide layer is used to compress and conduct incident light, enhance the interaction between light and the metal resonant cavities, and improve the selective transmittance of the target waveband; The integrated AI inference processing unit is coupled to a CMOS or CCD sensor, used to receive and decode in real time the encoded light intensity signals output by the sensor after being modulated by the array of tunable light filtering units, and reconstruct the continuous fine spectral curve corresponding to each pixel point.

[0012] To achieve the above purpose, the fourth aspect of the present application provides a sub-Angstrom synchronous spectral imaging method, comprising: S1, non-correlation encoding and modulation of incident light by an array of super surface resonant cavities, each resonant cavity independently designed to form a multi-dimensional spectral response without interference; S2, based on the light path compression structure of the integrated silicon nitride waveguide layer, synchronously transmitting the encoded spectral information to an image sensor to generate an encoded light intensity map; S3, using an FPGA carrying an AI inference unit to solve the encoded light intensity map in real time, and inversely reconstructing a spectral data cube with sub-Angstrom resolution according to the transmission characteristics of each resonant cavity; S4, combining the RGB image synchronously collected by a visible light camera, realizing cross-modal alignment and fusion of spatial-spectral information through a feature point matching algorithm.

[0013] The sub-Angstrom resolution low-altitude spectral remote sensing chip, system, device and method of the embodiment of the present application can solve the problem that the existing spectral remote sensing equipment loses subtle spectral differences in complex scenes due to insufficient resolution (≥0.3 nm) and cannot identify sub-Angstrom level molecular characteristics; the fixed compression strategy cannot adapt to the spectral complexity of the object, important information is lost or invalid data is redundant; mechanical scanning or data processing delay restricts the timeliness of imaging, and laboratory offline interpretation is delayed, which cannot meet the real-time needs of agricultural / disaster monitoring and is difficult to adapt to the unmanned aerial platform due to large volume and weight.

[0014] The present application has the advantages that, compared with the prior art, sub-Angstrom level hyperspectral resolution can be achieved, at the same time, mechanical scanning and data processing delay can be eliminated, and single-frame wide-field-of-view high-image-resolution spectral image acquisition can be achieved. The spectral camera, visible light observation camera, industrial computer and gimbal servo control are integrated in a single pod to realize multi-functional cooperation, the visible light camera returns real-time images to assist spectral positioning (such as locking the fire smoke diffusion track), the gimbal servo system adjusts the observation angle in response to the ground command, and the flexibility of remote sensing operation is significantly improved.

[0015] Additional aspects and advantages of the present application will be in part apparent and in part pointed out hereinafter. BRIEF DESCRIPTION OF DRAWINGS

[0016] The above and / or additional aspects and advantages of the present application will become apparent and be readily understood from the following description, taken in conjunction with the accompanying drawings, in which: Figure 1 is a structural diagram of a sub-Angstrom resolution low-altitude spectral remote sensing system provided by the embodiment of the present application; Figure 2 is a structural diagram of an integrated pod device based on the sub-Angstrom resolution low-altitude spectral remote sensing system provided by the embodiment of the present application; Figure 3 The embodiment of the present application provides a flowchart of a sub-Angstrom level synchronous spectral imaging method. DETAILED DESCRIPTION

[0017] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

[0018] In order to make the personnel in the technical field better understand the present application scheme, the technical scheme in the embodiment of the present application will be clearly and completely described below in combination with the drawings in the embodiment of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by the person skilled in the art without creative labor should belong to the scope of protection of the present application.

[0019] An ultralow-altitude spectral remote sensing system, device, chip and method according to an embodiment of the present application are described below with reference to the accompanying drawings.

[0020] Embodiment 1 Figure 1 An ultralow-altitude spectral remote sensing system according to an embodiment of the present application is shown in FIG. 1. Figure 1 As shown in FIG. 1, the ultralow-altitude spectral remote sensing system 10 comprises an optical imaging module 100, a spectral acquisition module 200, a storage communication module 300 and a comprehensive control module 400; wherein, The optical imaging module 100 is configured to capture spatial image information of a target area and pre-process incident light to provide high-quality incident light to the spectral acquisition module, while transmitting the split image or independently collected real-time image to the rear system to provide visual guidance for remote control of the remote sensing platform. The spectral acquisition module 200 is configured to decompose the incident light from the optical imaging module into a spectrum with sub-angstrom resolution and convert the spectrum into a corresponding electrical signal. The storage communication module 300 is configured to store the spectral data represented by the electrical signal output by the spectral acquisition module in real time, and perform bidirectional data interaction with the remote sensing platform, including flight parameters, system pose information, ground control instructions and data link state information. The comprehensive control module 400 is configured to coordinate the cooperative work of the optical imaging module, the spectral acquisition module and the storage communication module, and perform environmental adaptive compensation control, real-time data processing and servo control; wherein the environmental adaptive compensation control includes self-stabilization control, the real-time data processing includes image and information encoding, and the servo control is used to respond to ground instructions and drive the remote sensing platform to adjust the attitude.

[0021] It can be understood that the optical imaging module is responsible for the spatial information capture and light preprocessing of the target area, provides high-quality incident light for spectral light splitting, and can split or transmit real-time images to the rear for remote sensing control guidance; the spectral acquisition module decomposes the incident light into sub-angstrom resolution spectrum and converts it into an electrical signal; 3 storage communication module, real-time storage of spectral data and data interaction with the carrying platform, including flight information, device pose information, ground interactive control information, data link information, etc.; the integrated control module coordinates the work of each module, and performs environmental adaptive compensation (such as self-stabilization control), real-time data processing (image and information encoding), servo control (through ground instruction control remote sensing platform rotation) and other control functions.

[0022] In an embodiment of the present application, the optical imaging module is used to capture the spatial image information of the target area and preprocess the incident light to provide high-quality incident light to the spectral acquisition module, while transmitting the split image or independently collected real-time image to the rear system to provide visual guidance for remote control of the remote sensing platform; specifically, the module includes a front optical lens, a beam shaping component, and a light splitting system, the front optical lens is used to converge incident light from the target area, the beam shaping component performs uniformization, collimation and aberration correction processing on the incident light to improve the spatial consistency and energy distribution stability of the light field, ensuring that the light flux entering the spectral acquisition module is sufficient and the wavefront quality is excellent; the light splitting system uses a dichroic mirror or an optical fiber coupling method to divide the incident light into a visible light channel and a spectral channel according to the wavelength band: the visible light channel is directly transmitted to the integrated real-time imaging sensor to generate a high-frame-rate video stream for dynamic monitoring of the target area by ground operators and platform positioning; the spectral channel is guided to the spectral acquisition module for subsequent processing, realizing parallel operation of spatial imaging and spectral analysis, and improving the overall response efficiency of the system.

[0023] In an embodiment of the present application, the spectral acquisition module is used to decompose the incident light from the optical imaging module into a spectrum with sub-angstrom resolution, and convert the spectrum into a corresponding electrical signal; specifically, the module includes a miniature spectrometer core unit, which uses a snapshot spectral imaging technology based on a coded modulation chip, and through the integrated tunable filter unit array (composed of a silicon nitride waveguide layer and a metal resonant cavity) on the silicon substrate, the incident light is non-correlationally coded and modulated, realizing parallel acquisition of multiple spectral channels; each resonant cavity independently controls the transmittance of a specific wavelength band, breaking the crosstalk in the wavelength dimension in traditional spectral imaging, supporting single-exposure capture of full-view spectral information at a spectral resolution of 0.1 nm (i.e. sub-angstrom level); the coded and modulated light signal is received by the rear-end CMOS or CCD image sensor and converted into an electrical signal, outputting compressed sensing form coupled spectral data for subsequent real-time decoding to provide original data input.

[0024] In an embodiment of the present application, a storage communication module is used to store the spectral data represented by the electrical signals output by the spectral acquisition module in real time, and to interact with the remote sensing platform in both directions, including flight parameters, system pose information, ground control instructions and data link state information; Specifically, the module includes a high-speed solid-state storage unit and a multi-modal communication interface, the high-speed solid-state storage unit has large capacity and high write rate characteristics, and supports synchronous caching and persistent storage of high-dimensional spectral data stream, real-time visible light image and system metadata; the multi-modal communication interface supports Ethernet, optical fiber link or wireless data transmission link, and realizes real-time data exchange with the unmanned aerial vehicle flight control system, inertial navigation unit (IMU) and ground command station: on the one hand, flight parameters such as flight height, speed, attitude angle and platform pose information are received, which are used for subsequent geographic coordinate registration and motion compensation; on the other hand, spectral and image data, device state information are uploaded, and control instructions (such as acquisition mode switching, target area redirection, parameter adjustment, etc.) from the ground are received, to ensure that the system has good interactivity and response ability in complex task environment.

[0025] In an embodiment of the present application, a comprehensive control module is used to coordinate the cooperative work of the optical imaging module, the spectral acquisition module and the storage communication module, to perform environmental adaptive compensation control, real-time data processing and servo control; wherein the environmental adaptive compensation control includes self-stabilization control, the real-time data processing includes image and information encoding, and the servo control is used to respond to ground instructions and drive the remote sensing platform to adjust the attitude; Specifically, the module is composed of an embedded main control unit and a real-time operating system, and the timing operation of each module is coordinated through a unified time synchronization mechanism to ensure the timing consistency of data acquisition, transmission and processing; the environmental adaptive compensation control utilizes feedback information from the IMU and environmental sensors to implement image stabilization algorithms (such as electronic image stabilization or gimbal compensation), to suppress the influence of flight platform vibration and jitter on imaging quality; the real-time data processing function includes denoising, bad pixel correction and radiation calibration of the original image, rapid decoding and reconstruction of the coupled spectral data, and spatio-temporal alignment and fusion encoding of the visible light image and the spectral image to generate standardized data packets; the servo control function generates gimbal drive signals by analyzing ground instructions, controls the precise rotation of the pod in azimuth and pitch directions, realizes continuous tracking and multi-angle observation of dynamic targets, and comprehensively improves the intelligent level and task adaptability of the system.

[0026] Further, the overall workflow of the system is as follows: The entire system begins with the front-end light path modulation and coding process, which is the physical basis for realizing high-dimensional spectral information capture. The light of the target scene first enters the chip imaging system, is converged by the pre-optical system, and then irradiates the core structure of the chip - the silicon nitride waveguide layer integrated on the silicon substrate. The waveguide layer can effectively confine the incident light in the sub-wavelength micro-nano structure for transmission, playing a role in "compressing" the light path, not only improving the utilization efficiency of photons, but also creating conditions for the subsequent integration of miniaturized and high-density filter units. The light passing through the waveguide layer is guided to the metal resonant cavity array below, known as the "smart mask".

[0027] In the "smart mask" stage, the light undergoes key selective filtering and coding. Each micro metal resonant cavity in the array is precisely designed and has a unique spectral transmission characteristic, like an "optical fingerprint". Each resonant cavity only exhibits high transmittance for its pre-set specific waveband (target waveband), while strongly attenuating other wavebands. More importantly, the transmission spectra of all resonant cavities adopt a "non-correlation design", meaning they are completely independent and unrelated to each other in spectral response. Therefore, when the light passes through the array, the light irradiating each pixel on the image sensor below is no longer single-wavelength light, but a mixed light signal of multiple wavelengths modulated by a unique coding function, realizing the physical-level mixing and coding of spectral information.

[0028] Subsequently, the system enters the information capture stage. The mixed light signal modulated by coding is received by the CMOS or CCD image sensor located below the resonant cavity array. At this time, the sensor records not a traditional image, but a "coded image". In this image, the gray value of each pixel represents the weighted sum of the light intensity of all transmitted wavelengths after filtering by the corresponding resonant cavity. This two-dimensional coded image is essentially a compressed sensing result of a three-dimensional spectral data cube (space x, space y, wavelength λ) under a single exposure, capturing high-dimensional spectral information of the entire field of view in one shot, completely eliminating the time delay of traditional spectrometers relying on mechanical scanning.

[0029] Next, the back-end computational reconstruction stage uses integrated AI technology to decode the coded image in real time. The coded image is directly fed into the FPGA hardware platform integrated with the sensor, which is equipped with a dedicated AI inference unit. The AI unit has pre-stored the precise spectral response model of each resonant cavity, i.e., its "coding key". By running a deep learning decoding algorithm, the AI unit can inversely solve (decode) a complete and continuous spectral curve from the single intensity value received by each pixel of the sensor. Finally, the system reconstructs a three-dimensional data cube containing millions of pixels, each with fine spectral information, within milliseconds, realizing a complete closed loop from physical coding to digital decoding.

[0030] Finally, the system further enhances the analytical value of the data through cross-modal fusion. Simultaneously with spectral data acquisition, the system activates a high spatial resolution visible light camera to acquire RGB color images of the same scene. Since the two imaging methods may differ in viewpoint and resolution, the system employs advanced feature point matching algorithms such as SIFT to precisely align and register the hyperspectral and visible light images spatially. The aligned multimodal data (the "spectral fingerprint" of the hyperspectral image + the "spatial details" of the visible light image) is fused into a richer dataset, which can be directly input into subsequent AI application modules to perform more accurate material classification, environmental anomaly detection, and disease identification, significantly improving the accuracy and practicality of remote sensing interpretation.

[0031] The sub-angstrom resolution low-altitude spectral remote sensing system according to embodiments of the present invention, combined with optical imaging and spectral acquisition modules, provides real-time spatial positioning guidance for spectral acquisition, significantly shortening the decision-making time for spectral remote sensing. Its lightweight integration expands the application boundaries of low-altitude remote sensing, such as through unmanned aerial vehicle platforms, enhancing the flexibility of remote sensing operations.

[0032] Example 2 To achieve the above embodiments, this embodiment also provides an integrated pod device based on a sub-angstrom resolution low-altitude spectral remote sensing system. Based on the above system composition, the integrated pod device achieves high integration, small size, low weight, and high protection in its mechanical design, integrating electrical components such as a spectral acquisition camera, a visible light real-time observation camera, an industrial control computer, and a gimbal control unit to realize low-altitude remote sensing functions. For example... Figure 2 As shown: The device includes at least a central axis beam, a base, an optical cabin, and a comprehensive control board, wherein, The central axis beam extends longitudinally, with its upper end connected to the UAV platform via a base and its lower end connected to the optical cabin. The base integrates an azimuth angle sensor and an azimuth motor. The azimuth angle sensor is used to collect the azimuth angle information of the base relative to the UAV platform in real time and transmit it to the integrated control board. The azimuth motor is used to receive the drive commands issued by the integrated control board and drive the entire pod to rotate around the central axis beam, thereby adjusting the overall orientation of the pod. The optical cabin is located at the lower part of the pod, and the sub-angstrom spectral acquisition device, the visible light camera and the industrial computer are integrated inside; the sub-angstrom spectral acquisition device carries a sub-angstrom resolution low-altitude spectral remote sensing chip, and is used for collecting spectral data of the target area with high spectral resolution; the optical axis of the visible light camera is parallel or coaxial with the viewing axis of the sub-angstrom spectral acquisition device, and is used for synchronously collecting real-time visible light images of the target area to provide visual guidance for the image transmission of the unmanned aerial vehicle remote control operation; the industrial computer is in communication connection with the unmanned aerial vehicle platform and the comprehensive control board, and is used for receiving control instructions, coordinating and controlling the working states of the sub-angstrom spectral acquisition device and the visible light camera, and storing and forwarding the collected spectral data and visible light images in real time; The comprehensive control board is integrated in the optical cabin, and is in bidirectional data communication with the flight control system and the image transmission system of the unmanned aerial vehicle platform through an open interface, receives flight parameters, control instructions and time synchronization signals from the unmanned aerial vehicle; the comprehensive control board also connects and processes the feedback signals of the pitch angle sensor, the pitch speed sensor and the azimuth speed sensor arranged in the optical cabin, and solves the complete pose information of the optical cabin in real time; based on the pose information and the received control instructions, the comprehensive control board executes a servo control algorithm, generates a driving signal and outputs it to the azimuth motor and the pitch motor in the optical cabin, realizes stable pointing and dynamic tracking of the optical cabin in two degrees of freedom of azimuth and pitch, and completes stable imaging and remote sensing detection of the target area.

[0033] Table 1 is the name of each module of the device, as shown in Table 1: Table 1

[0034] Specifically, according to the structure of the system block diagram, the entire device is connected from top to bottom by the central axis bundle, realizing the mechanical support, electrical interconnection and information interaction between the unmanned aerial vehicle platform and the optical load. The upper end of the central axis bundle is firmly connected to the unmanned aerial vehicle body through a highly integrated base, which not only bears the weight of the entire pod and the dynamic load during flight, but also has built-in key azimuth control components. The inside of the base is integrated with a high-precision azimuth angle sensor (such as a magnetic encoder or an optical encoder) and an azimuth motor (such as a brushless direct-drive motor). The azimuth angle sensor collects real-time absolute azimuth angle information of the base relative to the unmanned aerial vehicle body and transmits the data to the integrated control board located in the optical cabin through the wire bundle; the azimuth motor receives the closed-loop control instructions generated from the integrated control board, drives the base to rotate continuously around the central axis bundle by 360°, thereby realizing large-range and high-precision azimuth pointing adjustment of the pod as a whole in the horizontal direction. At the same time, the base communicates with the flight control system, the image transmission system, the power management system and other core modules of the unmanned aerial vehicle through the open interface (such as CAN bus, Ethernet or serial communication interface) reserved by the unmanned aerial vehicle, realizes real-time interaction of flight parameters (such as position, speed, attitude), time synchronization signal and remote control instructions, and ensures high coordination between pod control and flight platform.

[0035] The optical cabin is located at the lower end of the central axis bundle and is the sensing core of the entire system, which is integrated with sub-angstrom spectral acquisition equipment, a visible light camera, an industrial computer, an integrated control board and a servo drive mechanism. The sub-angstrom spectral acquisition equipment is the core sensor of the optical cabin, which is equipped with a sub-angstrom resolution low-altitude spectral remote sensing chip based on coding modulation technology, which can capture high-dimensional data cubes containing rich spectral features with a spectral resolution of 0.1 nm, which can be used for subsequent fine material identification and analysis. The visible light camera is coaxial or parallel with the sub-angstrom spectral acquisition equipment, and their optical axes are completely consistent, which is used to synchronously acquire high-resolution and high-frame-rate RGB visible light images of the target area. The image not only provides intuitive visual information for ground operators, but also serves as a key image transmission source for unmanned aerial vehicle remote control and target positioning, assisting in realizing precise flight control and target locking. The industrial computer, as a data processing and storage center, receives unified scheduling instructions from the integrated control board, is responsible for coordinating the start, parameter configuration and working time sequence of the sub-angstrom spectral acquisition equipment and the visible light camera, and performs real-time storage, compression and forwarding of the massive spectral data and high-definition visible light video stream collected, and transmits them back to the ground station through the wireless data link.

[0036] The integrated control board is the "brain" of the entire pod, responsible for closed-loop control and intelligent decision-making of the system. It receives remote control commands and flight status information from the UAV flight control system, and fuses data from various sensors inside the optical pod, including pitch angle sensors (such as inclinometers or gyroscopes), pitch speed sensors (such as encoder feedback), and azimuth angle information from the base. It calculates the complete pose (attitude angle and angular velocity) of the optical pod in three-dimensional space in real time. Based on this information, the integrated control board runs advanced servo control algorithms (such as PID or adaptive control) to generate precise motor drive signals. These signals are sent not only to the azimuth motor of the base, but also to the pitch motor inside the optical pod, driving the optical pod to swing up and down in the pitch direction, thus achieving accurate pointing and dynamic tracking of the target area in two dimensions (azimuth-pitch). More importantly, the integrated control board has the ability to adapt to the environment, actively adjusting servo control parameters to drive the motor to compensate in the opposite direction, ensuring that the optical pod remains stable in complex flight environments, effectively suppressing image blur, and ensuring high-quality hyperspectral and visible light imaging. Through this multi-level, multi-sensor fusion closed-loop control architecture, the system ultimately achieves full-process automation and intelligence from command reception, pose perception, servo drive to stable imaging, meeting the stringent requirements of low-altitude remote sensing for high precision and high stability observation.

[0037] Embodiment 3 To achieve the above embodiment, the embodiment also provides an encoding and modulation chip for sub-Angstrom spectral image snapshot acquisition, comprising a silicon substrate, an array of tunable filter units integrated on the silicon substrate, and an integrated AI inference processing unit. The array of tunable filter units is composed of a silicon nitride waveguide layer and an array of metal resonant cavities stacked on the silicon substrate. Each metal resonant cavity serves as an independent filter unit, achieving a non-correlation designed spectral transmittance response by adjusting its physical structure, physically encoding and modulating the incident light to break the spectral correlation between pixels in traditional imaging, thereby simultaneously capturing multi-dimensional spectral information with a resolution of 0.1 nm under a single exposure. The silicon nitride waveguide layer is used to compress and conduct incident light, enhance the interaction between light and metal resonant cavities, and improve the selective transmittance of the target waveband. The integrated AI inference processing unit is coupled to a CMOS or CCD sensor for receiving and decoding in real time the encoded light intensity signal output by the sensor after being modulated by the array of tunable filter units, and reconstructing the continuous fine spectral curve corresponding to each pixel point.

[0038] It can be understood that the chip of the embodiment uses an encoding modulation scheme to realize snapshot acquisition of sub-angstrom level spectral images, which mainly realizes high-dimensional spectral information acquisition through miniaturization and integrated mask, and the principle is to use metasurface technology to stack silicon nitride waveguide layer and metal resonant cavity array on a silicon substrate to form a tunable light filtering unit. Each resonant cavity independently modulates the light transmittance (non-correlation design), breaks the spectral correlation between traditional pixels, and realizes the synchronous capture of multi-dimensional spectrum with a resolution of 0.1 nm. The waveguide layer compresses the light wavelength, and the resonant cavity enhances the transmittance of the target waveband through resonance. Different designs of the mask can realize a heterogeneous mask system, and the mask characteristics can be adjusted according to different spectral high-resolution and spatial resolution requirements. Through the FPGA period carrying an integrated AI inference unit, the spectral coupling information collected by the CMOS or CCD sensor under the mask can be realized. Hardware-level operation, real-time analysis of non-correlated spectral data. Through the chip of the application, cross-modal alignment can also be performed, and visible light cameras and spectral imaging modules can be used for simultaneous acquisition. The SIFT algorithm can be used to match the feature points of the visible light and spectral images, and support can be provided for subsequent decision fusion.

[0039] Specifically, the chip of the embodiment uses an encoding modulation scheme to realize snapshot acquisition of sub-angstrom level spectral images, which mainly realizes high-dimensional spectral information acquisition through miniaturization and integrated mask, and the principle is to use metasurface technology to stack silicon nitride waveguide layer and metal resonant cavity array on a silicon substrate to form a tunable light filtering unit. The light filtering unit as a core coding mask is integrated in front of the image sensor to realize real-time information modulation in the optical domain. Each metal resonant cavity as an independent spectral modulation unit, its geometric structure (such as size, shape, period) is optimized and designed, which can produce localized surface plasmon resonance (LSPR) or Fabry-Perot resonance effect in a specific wavelength range, thereby selectively enhancing the transmittance of the target waveband and suppressing the non-target waveband; more importantly, the light transmittance response mode of each resonant cavity adopts a non-correlation coding strategy, that is, the spectral transmittance functions of adjacent units are not related in the wavelength dimension, which completely breaks the spectral redundancy and crosstalk between pixels caused by the filter array or prism in traditional spectral imaging, and realizes the decoupling of spatial-spectral dimensions. On this basis, the chip can simultaneously capture high-dimensional spectral data cubes in the full field of view under single exposure (snapshot mode), with a spectral resolution of 0.1 nm (sub-angstrom level), which is significantly better than traditional push-broom or filter wheel spectral instruments, and is suitable for real-time remote sensing monitoring of dynamic targets.

[0040] Further, as the basic structure for light field regulation, the silicon nitride waveguide layer not only has the characteristics of wide band (visible to near infrared) and low loss transmission, but also can efficiently focus the incident light energy to the underlying resonant cavity array through mode compression effect, improving the photon utilization rate and signal-to-noise ratio. The waveguide layer and the resonant cavity are coupled through a nanoscale gap to further enhance the light-matter interaction strength and improve the selective response of specific wavebands. By precisely controlling the thickness, refractive index distribution of the waveguide layer and the arrangement density of the resonant cavity, fine wavelength compression and spatial modulation of the incident light can be realized, providing high-quality physical coding input for high-resolution spectral reconstruction.

[0041] Further, by designing different masks, a heterogeneous coding mask system can be realized. For different application scenarios with different requirements for spectral resolution and spatial resolution, the mask characteristics can be flexibly adjusted. For example, in the mineral identification task requiring ultra-high spectral resolution, a resonant cavity array with dense and narrowband response peaks can be designed. In the vegetation health monitoring task requiring high spatial resolution, a sparse and broadband response coding mode can be used to improve the spatial sampling density. In addition, a reconfigurable mechanism (such as integrated micro-heater or electro-optic material) can be introduced to dynamically tune the response characteristics of the resonant cavity, making the same chip adapt to multiple tasks and multiple environments for remote sensing requirements, significantly improving the adaptability and reuse value of the system.

[0042] Further, by loading an FPGA or an application-specific integrated circuit (ASIC) integrated with an AI inference unit, the two-dimensional coupled spectral images (i.e., compressed sensing measurement results) collected by the CMOS or CCD sensor under mask modulation can be processed to realize hardware-level parallel operation and real-time analysis of non-correlated spectral data. The AI inference unit is embedded with a lightweight deep learning decoding model (such as a convolutional neural network or a U-Net variant). The model learns the nonlinear mapping relationship between the encoding matrix of the mask and the real spectrum in the training stage, and can complete high-quality reconstruction from the coupled image to the three-dimensional spectral cube within milliseconds after deployment, avoiding the high computational overhead and delay caused by traditional iterative algorithms. The entire decoding process is completed locally on the chip, without the need for external computing resources, meeting the stringent requirements of real-time performance and low power consumption in edge computing scenarios such as airborne and satellite.

[0043] Finally, through the chip of the application, cross-modal alignment can also be performed, synchronous acquisition is performed using the integrated visible light camera and spectral imaging module, and time and space consistency of multi-modal data is realized. Specifically, the chip supports a dual-channel parallel acquisition mode: one channel acquires high-resolution spectral images through an encoding mask, and the other channel acquires high-frame-rate and high-spatial-resolution RGB images through a bypass visible light sensor. The two channels of data are strictly synchronized in time stamp and spatial field of view, and subsequently, key feature points in the two images can be automatically extracted and matched through SIFT, SURF or deep feature matching algorithms, to complete pixel-level geometric registration and spatial alignment. This function provides an accurate spatial reference for subsequent multi-modal decision fusion (such as superimposing spectral classification results on visible light images, and visualizing abnormal area labeling), significantly improves the accuracy and interpretability of remote sensing interpretation, and is widely used in complex task scenarios such as environmental monitoring, precision agriculture and disaster assessment.

[0044] It can be known that the sub-Ermi resolution spectral remote sensing chip integrates spectral modulation and sub-Ermi resolution snapshot information acquisition capability, and can realize wide field of view spectral imaging without mechanical scanning. High spectral resolution and high spatial resolution remote sensing data acquisition are realized at the same time, the acquisition efficiency is greatly improved, and the form of the industry spectral remote sensing is overturned. Combined with optical imaging and spectral acquisition module, real-time spatial positioning guidance is provided for spectral acquisition, and the spectral remote sensing decision time is greatly shortened. The lightweight integration expands the boundaries of low-altitude remote sensing applications such as unmanned aerial vehicle platforms, and improves the flexibility of remote sensing operation.

[0045] In summary, the core of this chip lies in its revolutionary design at the physical level, which is based on a unique array of filter units. Through advanced metasurface micro-nanofabrication technology, the chip integrates an array of millions of micro metal resonant cavities on a silicon substrate, each serving as an independent filter unit, whose physical structure (including shape, size, periodic arrangement, and material combination) can be precisely controlled. This high-precision manufacturing capability enables each filter unit to have a tailor-made, unique spectral transmittance curve. For example, one unit may have a high and narrow transmission peak for 500 nm wavelength light, its adjacent unit may have a wide and gentle response to 502 nm light, and another unit may simultaneously enhance the transmission of two wavelength bands of 450 nm and 650 nm. This non-correlation, heterogeneous design is equivalent to providing each pixel of the image sensor with a dedicated miniature spectrometer, completely overturning the traditional imaging mode of millions of pixels sharing a limited number of filters (such as RGB), and achieving a leap in spectral sensing capability from the hardware bottom. In terms of information acquisition mechanism, this chip realizes the paradigm shift from traditional "selection" to "encoding". The working mode of traditional filters is "selective transmission", for example, a red filter only allows red light to pass through, while directly blocking or absorbing other wavelengths of light, resulting in the permanent loss of most spectral information at the physical level. The filter units of this chip are not used for "selection", but as "encoders". It allows light of different wavelengths to pass through at the same time with a specific, known weight ratio, eventually forming a weighted mixed light intensity value containing multi-band information on the sensor pixel. This single intensity value does not directly correspond to a certain color or wavelength, but is the "ciphertext" of the original spectral information after physical encryption, thus preserving the high-dimensional spectral characteristics of the incident light in a single exposure, avoiding the resolution bottleneck caused by information loss in traditional methods. At the information restoration level, this chip abandons the traditional "guessing" method relying on interpolation, and instead adopts the "equation solving" method based on prior knowledge. Since the traditional method loses spectral information, subsequent processing can only use simple interpolation of the values of adjacent pixels to "guess" the missing bands, with limited accuracy. The decoding process of this chip is completely different: since the precise spectral response curve (i.e. "encryption key") of each resonant cavity is known at the time of manufacture and pre-stored in the system, the task of the integrated AI inference unit (such as FPGA) is to use this prior knowledge to treat each pixel's mixed light intensity value captured by the sensor as a linear equation about the incident spectrum. By simultaneously solving millions of such equations (i.e. a large-scale inversion problem), the AI unit can accurately and in real time calculate the complete and continuous spectral curve of each pixel, thus achieving high-quality reconstruction from the encoded signal to a high-resolution spectral data cube.

[0046] Example 4 Figure 3 is a flow chart of the sub-Angstrom synchronous spectral imaging method of one embodiment of the present application.

[0047] As shown in Figure 3 The sub-Angstrom synchronous spectral imaging method comprises the following steps: S1, non-correlation coding modulation of incident light by a super surface resonant cavity array, each resonant cavity independently designed to form a multi-dimensional spectral response without interference.

[0048] Specifically, this step is based on metasurface technology, by integrating a silicon nitride waveguide layer and a metal resonant cavity array on a silicon substrate, a miniature spectral modulation unit with high designability and independent response characteristics is constructed. Each resonant cavity unit is independently designed by its geometric structure (such as nanocolumn, open ring, fishbone structure, etc.) and material parameters (such as metal type, thickness, dielectric constant, etc.), so as to realize high transmittance response to specific waveband, and significant suppression or attenuation characteristics to other wavebands.

[0049] Specifically, the transmission curve of the resonant cavity is determined by its resonance frequency, which can be precisely controlled by changing the size, period, filling factor, etc. of the cavity. In this patent, the resonant cavity corresponding to each pixel has mutually independent transmission functions, i.e. their transmission spectra are orthogonal or approximately orthogonal to each other in the wavelength domain, thereby realizing parallel coding of multi-dimensional spectral information at the physical level. This non-correlation design breaks the linear correlation of spectral response between pixels in traditional spectral imaging, so that the output light intensity of each pixel point is the weighted sum of multiple wavelength components, rather than the direct measurement of a single waveband.

[0050] The transmission peak width of each resonant cavity in this step can be controlled within 0.1 nm, and the center wavelength coverage range is 400 nm to 1000 nm, meeting the requirements of sub-Angstrom spectral resolution. At the same time, the total number of resonant cavities in the array can reach tens of thousands, and the transmission curve of each unit is optimized by full-wave electromagnetic simulation (such as FDTD or CST), ensuring good signal-to-noise ratio and dynamic response range (≥60 dB) in a wide waveband range.

[0051] The technical effect of this step is that, by non-correlation coding modulation, high-dimensional parallel capture of spectral information is realized, snapshot spectral imaging can be completed without mechanical scanning, greatly improving the imaging efficiency and real-time performance of the system. At the same time, this design effectively avoids the time error caused by scanning delay or interference pattern processing in traditional spectrometers, providing a high-quality, high-density raw data basis for subsequent AI decoding and cross-modal fusion.

[0052] Further, S1 comprises: S11, a resonant cavity with a metal-insulator-metal structure, achieves transmission peaks with a wavelength interval of 0.1 nm by adjusting the period, width, and depth parameters of the resonant cavity.

[0053] Specifically, the resonant cavity with a metal-insulator-metal (MIM) structure achieves transmission peaks with a wavelength interval of 0.1 nm by adjusting its period, width, and depth parameters, which is one of the key steps to achieve sub-angstrom spectral resolution. This step is based on the collaborative design of metasurfaces and nanophotonic structures, and by precisely controlling the geometric parameters of the resonant cavity, high-selectivity modulation and coding of incident light waves are achieved.

[0054] In some implementations, the MIM resonant cavity is composed of two layers of metal thin films (such as gold or silver) sandwiching a layer of high dielectric constant insulating material, forming an electromagnetic resonant structure in the vertical direction. Its working principle is based on the Surface Plasmon Resonance (SPR) effect, when the incident light wavelength matches the structure parameters of the resonant cavity, a high-transmission resonance peak will be generated at a specific wavelength. By changing the period (P), width (W), and depth (D) of the resonant cavity, its resonant wavelength and bandwidth can be adjusted, thereby achieving a transmission peak arrangement with a wavelength interval of 0.1 nm.

[0055] Specifically, the period P is usually in the range of 200-400 nm, the width W is between 50-200 nm, and the depth D is between 50-150 nm. The combination of these parameters needs to meet specific electromagnetic coupling conditions to ensure that the wavelength interval between adjacent resonant peaks is stable at 0.1 nm. In the design process, finite element method (FEM) or finite difference time domain method (FDTD) is used for electromagnetic simulation to optimize the structure parameters to achieve the required spectral response characteristics. In addition, to enhance the contrast and signal-to-noise ratio of the resonant peak, a silicon nitride waveguide layer is optionally introduced below the resonant cavity to enhance the transmission efficiency of the target waveband through the coupling of waveguide modes and resonant cavities.

[0056] This step plays a key role in the front-end spectral modulation of the system, and its high-density, non-correlation design enables each pixel point to independently capture spectral information of different wavelengths, thereby achieving simultaneous acquisition of multi-dimensional spectral data in a single frame image. Compared with traditional gratings or interferometric spectrometers, this method does not require mechanical scanning, greatly improving the imaging speed and system stability, while still maintaining high light flux and signal-to-noise ratio at sub-angstrom resolution, providing a high-quality raw data foundation for subsequent AI decoding and cross-modal fusion.

[0057] S12, by utilizing the coupling effect of the silicon nitride waveguide layer and the metal resonant cavity, the refractive index distribution of the waveguide layer is changed to optimize the steepness of the transmission rate of the resonant cavity, so that the signal separation degree of adjacent wavebands is ≥95%.

[0058] Specifically, in some implementations, the present application optimizes the steepness of the resonant cavity transmittance by utilizing the coupling effect between the silicon nitride waveguide layer and the metal resonant cavity, thereby improving the separation degree of adjacent band signals to ≥95%. The technical implementation of this step is based on the waveguide-resonant cavity coupling structure in a photonic integrated circuit (PIC), and the core principle is to change the resonance condition of the resonant cavity by adjusting the refractive index distribution of the waveguide layer, thereby enhancing its wavelength selectivity.

[0059] Specifically, the silicon nitride waveguide layer has a high refractive index contrast (n≈2.0) and low loss (<0.1 dB / cm), which can efficiently guide incident light to the metal resonant cavity array. The metal resonant cavity is usually made of high-reflectivity materials such as gold (Au) or silver (Ag), and its structure can be a plasmonic resonator or a metamaterial resonator, with a working wavelength range of 400-1000 nm. By introducing a refractive index gradient distribution in the waveguide layer (such as using electron beam lithography and reactive ion etching technology to form a periodic refractive index modulation in the waveguide), the coupling efficiency of the resonant cavity can be dynamically adjusted. For example, when the refractive index of the waveguide layer is lowered in a specific region (such as from 2.0 to 1.8), the resonant wavelength of the resonant cavity can be shifted, thereby enhancing the steepness of its transmission spectrum (Full Width at Half Maximum, FWHM<0.05 nm).

[0060] In practical applications, this step can be integrated into the front-end optical module of a silicon-based spectral remote sensing chip for sub-angstrom level wavelength selective modulation of incident light. By optimizing the refractive index distribution of the waveguide, the system can achieve high-resolution, snapshot spectral acquisition without increasing mechanical scanning or complex interference structures. This technical means effectively solves the contradiction between resolution and throughput in traditional gratings and Fourier transform spectrometers, providing key support for dynamic target identification in unmanned aerial remote sensing, such as pollution diffusion and vegetation health monitoring.

[0061] S2, based on the optical path compression structure integrated with the silicon nitride waveguide layer, synchronously transmits the encoded spectral information to the image sensor to generate an encoded light intensity map.

[0062] Specifically, in some implementations, based on the optical path compression structure integrated with the silicon nitride waveguide layer, the encoded spectral information is synchronously transmitted to the image sensor to generate an encoded light intensity map, which is one of the core steps of the present application to achieve sub-angstrom level hyperspectral resolution remote sensing. This step combines miniaturized optical structures with non-correlation spectral coding strategies to achieve parallel acquisition and compressed transmission of spectral information in the spatial dimension.

[0063] The structure uses a stacked silicon nitride waveguide layer on a silicon substrate as an optical path compression medium. Silicon nitride is widely used in integrated photonic devices due to its low loss, high refractive index contrast, and good thermal stability. The waveguide layer is formed by micro-nano processing technology, with a width of usually between 200-400 nm, a height of about 100-200 nm, and a refractive index of about 2.0. The incident light is constrained to propagate in the waveguide, achieving spatial and wavelength dimension compression, thereby reducing the optical path length and improving system integration. Subsequently, the optical signal enters the encoding mask array composed of metal resonant cavities, each resonant cavity having an independently designed resonance wavelength and bandwidth, usually within the range of 0.1-0.5 nm, and the spectral responses of each other are not related, ensuring that the light signal received by each pixel point is a multi-band weighted mixed encoded light intensity.

[0064] The supported spectral range of the structure is 400-1000 nm, with adjustable central wavelength and controlled bandwidth within 0.1 nm, meeting the sub-angstrom resolution requirement. The encoding mask array is usually composed of 1024x1024 or higher pixels, each pixel corresponding to an independent resonant cavity, and its transmittance function is precisely controlled by design parameters such as cavity size, period, and material thickness. In the light intensity acquisition stage, the CMOS or CCD image sensor records the total light intensity value of each pixel with a 12-16 bit dynamic range, forming an encoded light intensity map, whose data volume can be compressed to less than 1 / 10 compared to the traditional wave-by-wave scanning method.

[0065] This step is suitable for rapid identification of ground material composition in unmanned aerial vehicle low-altitude remote sensing systems, such as crop disease and pest detection in agriculture, pollutant diffusion tracking in environmental monitoring, etc. By synchronously transmitting the encoded light intensity map, the system can capture the complete spectral information of the target area in a single image without mechanical scanning, significantly improving the imaging efficiency and real-time performance.

[0066] Further, S2 includes: S21, a three-dimensional optical path compression network is formed by multiple layers of silicon nitride waveguides, and the refractive index difference between the waveguide layers is used to control the light propagation path, so that the spectral information forms a two-dimensional spatial distribution on the sensor plane.

[0067] Specifically, in this embodiment, "a three-dimensional optical path compression network is formed by multiple layers of silicon nitride waveguides, and the refractive index difference between the waveguide layers is used to control the light propagation path, so that the spectral information forms a two-dimensional spatial distribution on the sensor plane" is the core optical structure design step for achieving sub-angstrom level high spectral resolution remote sensing. This step uses the stacking and refractive index engineering of multiple layers of silicon nitride waveguides to construct a three-dimensional optical path compression network, thereby achieving spatial encoding and efficient capture of spectral information without relying on mechanical scanning.

[0068] In some implementations, the three-dimensional light path compression network is composed of several layers of silicon nitride waveguides, each layer of waveguide has a different refractive index distribution, and the coupling and propagation path of light between different waveguide layers is controlled by the interlayer refractive index difference (Δn). Specifically, the silicon nitride waveguide layer is usually deposited on a silicon-based substrate, and the PDK (Process Design Kit) standard process is adopted, the waveguide width is between 400-600 nm, the height is 100-200 nm, and the interlayer refractive index difference Δn is controlled in the range of 0.05-0.15 to realize the mode conversion and spatial shift of light of a specific wavelength. By accurately designing the geometric parameters and refractive index distribution of the waveguide, the mode interference and phase modulation of light of different wavelengths can occur between the waveguide layers, thereby forming a two-dimensional light intensity distribution with wavelength dependence on the sensor plane.

[0069] Further, the three-dimensional waveguide structure is combined with an array of metal resonant cavities to form a tunable spectral encoding mask. Each resonant cavity has an independent resonance wavelength, and the coupling efficiency is controlled by the refractive index difference of the waveguide layer to realize non-correlation modulation of the incident light. In the wavelength compression process, the spectral information is mapped into a two-dimensional pixel array of the sensor to form a high-dimensional spectral encoding image. This design can support a spectral resolution of 0.1 nm while maintaining a high spatial resolution (≥1000×1000 pixels), meeting the needs of fine identification of ground objects in low-altitude remote sensing.

[0070] This step plays a key role in the front-end spectral modulation of the system, providing high-quality, high-dimensional raw data for subsequent AI decoding and spectral cube reconstruction. The innovation lies in realizing light path compression and wavelength selective modulation through the refractive index difference between waveguide layers, breaking through the physical limitations of traditional gratings and interferometers in resolution and flux, and providing a feasible optical architecture foundation for realizing lightweight, snapshot high-spectral imaging.

[0071] S22, a graded refractive index waveguide design is adopted to produce a preset angle shift of light of different wavebands when propagating in the waveguide, thereby realizing orthogonal separation of spectral dimension and spatial dimension.

[0072] Specifically, in some implementations, the present application adopts a graded refractive index waveguide design to realize a preset angle shift of light of different wavebands when propagating in the waveguide, thereby realizing orthogonal separation of spectral dimension and spatial dimension. This technology is based on the transverse gradient distribution of the refractive index in the waveguide, and by adjusting the refractive index function of the waveguide material, different wavelengths of light will have an angle shift due to the refractive index difference during propagation, and finally form spatially distributed spectral information on the detector.

[0073] In practical applications, the waveguide structure is integrated on a silicon-based chip to form a tunable filtering unit with an array of metal resonant cavities. Each resonant cavity has an independent resonance wavelength that cooperates with the refractive index gradient of the waveguide to further enhance the transmission efficiency of a specific waveband. The design is suitable for unmanned aerial vehicle low-altitude remote sensing scenarios, especially in applications that require rapid and high-precision acquisition of ground object spectral information, such as farmland heavy metal pollution detection, vegetation health assessment, etc.

[0074] S3, using the FPGA equipped with an AI inference unit to perform real-time calculation on the encoded light intensity map, and inversely deducing a spectral data cube with sub-angstrom resolution based on the transmission characteristics of each resonant cavity.

[0075] Specifically, in the spectral data reconstruction step of the present application, the FPGA equipped with an AI inference unit is used to perform real-time calculation on the encoded light intensity map, and inversely deduce a spectral data cube with sub-angstrom (0.1 nm) resolution based on the transmission characteristics of each resonant cavity, which is the core step of realizing high-precision and snapshot spectral imaging. This step is based on the principle of non-correlation spectral coding, combining hardware acceleration and deep learning algorithms to complete efficient mapping from physical coding to spectral decoding.

[0076] This step first relies on the transmission function of each independent resonant cavity in the front-end mask. These resonant cavities are composed of silicon nitride waveguides and metal resonant structures on a silicon substrate, and their transmission spectra are accurately modeled through electromagnetic simulation (such as the FDTD method) during the design phase, ensuring that the spectral response of each pixel has a high degree of non-correlation. During imaging, the incident light is modulated by the mask to form an encoded light intensity map that is a weighted superposition of multiple non-correlated spectral response functions. This map is captured by a CMOS or CCD image sensor at a high frame rate (≥30 fps). FPGA, as a real-time processing platform, has an AI inference unit (such as an embedded neural network based on TensorFlow Lite or ONNX) built-in, which can perform parallel calculation on the encoded image.

[0077] The processing capability of FPGA needs to meet the real-time decoding requirements of at least 1000 independent resonant cavities, and the spectral calculation of each pixel needs to be completed within 10ms to support an imaging rate of ≥30fps. The AI model input is an N×N pixel encoded light intensity matrix, and the output is a continuous spectral curve corresponding to each pixel (such as in the range of 350-1000nm, with a wavelength point every 0.1nm), and the calculation accuracy needs to reach ±0.01nm to meet the sub-angstrom resolution requirement. In addition, the system needs to support dynamic adjustment of the calculation weight to adapt to the spectral complexity of different ground objects.

[0078] This step can be widely applied to unmanned aerial vehicle low-altitude remote sensing tasks such as farmland heavy metal pollution monitoring, mineral resource identification, urban heat island effect analysis, etc. During flight, the system assists positioning through visible light images, ensures that the spectral data accurately correspond to the geographic coordinates, and thus realizes high spatiotemporal resolution remote sensing data fusion.

[0079] Further, S3 comprises: S31, performing convolutional neural network algorithm by FPGA hardware acceleration to perform nonlinear inversion on the coded light intensity value of each pixel point to generate a continuous spectrum curve.

[0080] In the technical scheme of the application, this step is the core link for realizing real-time decoding and reconstruction of high-dimensional spectral information. Specifically, by integrating a convolutional neural network (CNN) algorithm in an FPGA (field programmable gate array), the nonlinear inversion of the pixel light intensity value coded by the non-correlation resonant cavity mask is performed, thereby recovering the continuous spectrum curve corresponding to each pixel point. This process breaks through the limitation of traditional spectrometers relying on linear interpolation or Fourier transform, and realizes high-precision, low-delay spectral data processing.

[0081] FPGA, as a hardware acceleration platform, has parallel computing capability and low power consumption characteristics, and is particularly suitable for processing high-dimensional, real-time spectral data. The CNN model is deployed in the FPGA, and the input of the model is the coded light intensity matrix collected by the CMOS or CCD image sensor, each pixel point corresponding to a light intensity value modulated by the mask. The model extracts features and performs nonlinear mapping on the input data through multiple convolution kernels, wherein the number of channels of the convolution layer matches the number of target spectral bands, and is usually set to 1000-2000 channels to cover the 350-1000 nm band range, and each channel corresponds to a spectral resolution of 0.1 nm. The activation function uses a nonlinear function such as ReLU or Swish to enhance the model's fitting ability for complex spectral responses. Finally, the continuous spectrum curve of each pixel point is output through the fully connected layer, usually 512-1024 wavelength points of intensity distribution.

[0082] The input size of the CNN model is usually 512x512 pixels, and each pixel point contains a real intensity value. The training data of the model is based on the accurate transmittance curve of the resonant cavity mask (obtained by simulating the nanoscale super surface structure), and the mapping relationship between the coded light intensity and the original spectrum is established through inverse problem modeling. The clock frequency of the FPGA is set to 200-500 MHz to ensure that the processing time of a single frame of image is controlled within 10-30 ms, meeting the real-time requirements of unmanned aerial vehicle low-altitude remote sensing. In addition, the quantization precision of the model is usually 8-16-bit fixed point number to balance between calculation accuracy and hardware resources.

[0083] The FPGA hardware accelerates the CNN algorithm to realize real-time decoding of high-resolution spectral data, and solves the contradiction between processing speed and accuracy of the traditional spectrometer, thereby providing an efficient and intelligent spectral processing capability for the low-altitude remote sensing system of the unmanned aerial vehicle platform.

[0084] In the spectral decoding step of the present application, the orthogonal matching pursuit (OMP) algorithm is used to perform sparse decoding on the transmission matrix encoded by the resonant cavity array, thereby realizing high-precision and low-latency spectral image reconstruction. This step is a key link in the entire system from physical encoding to spectral information restoration. The technical implementation is based on the compressed sensing theory (CS), and through iterative optimization strategy, the calculation complexity is significantly reduced while ensuring the spectral resolution of sub-angstrom level (0.1 nm), and the decoding delay is controlled within ≤5 ms, meeting the needs of real-time remote sensing data processing of the low-altitude unmanned aerial vehicle platform.

[0085] In the spectral decoding step of the present application, the orthogonal matching pursuit (OMP) algorithm is used to perform sparse decoding on the transmission matrix encoded by the resonant cavity array, thereby realizing high-precision and low-latency spectral image reconstruction. This step is a key link in the entire system from physical encoding to spectral information restoration. The technical implementation is based on the compressed sensing theory (CS), and through iterative optimization strategy, the calculation complexity is significantly reduced while ensuring the spectral resolution of sub-angstrom level (0.1 nm), and the decoding delay is controlled within ≤5 ms, meeting the needs of real-time remote sensing data processing of the low-altitude unmanned aerial vehicle platform.

[0086] The OMP algorithm selects the atom (i.e. the transmission response function of the resonant cavity) most related to the current residual in each iteration to gradually approach the original spectral signal. The number of iterations of the OMP algorithm is usually controlled within 5-10 times to balance between accuracy and speed. To meet the decoding delay of ≤5 ms, the algorithm is hardware accelerated on the FPGA, adopts parallel processing structure, combines fixed-point operation and pipeline design, and compresses the single iteration time to about 0.5 ms. In addition, the preprocessing accuracy of the dictionary matrix Φ needs to reach ±0.01 dB to ensure the accuracy of the spectral response function in the decoding process, thereby improving the signal-to-noise ratio (SNR) of the reconstructed spectrum to ≥30 dB.

[0087] The decoding step is suitable for real-time spectral analysis in unmanned aerial vehicle low-altitude remote sensing tasks, such as farmland heavy metal pollution detection, mineral composition identification, urban heat island effect monitoring, etc. In these scenarios, the target spectral features usually have sparsity, and the OMP algorithm can efficiently extract key band information, avoiding the mechanical delay and calculation bottleneck caused by traditional wave-by-wave scanning or Fourier transform.

[0088] Through the sparse decoding strategy, the accuracy and speed of spectral reconstruction are improved, and the dependence on high-bandwidth communication and large-capacity storage is effectively reduced, thereby providing core support for realizing a lightweight, low-power, and high-real-time unmanned aerial vehicle spectral remote sensing system.

[0089] S4, combined with the RGB image synchronously collected by the visible light camera, realizes cross-modal alignment and fusion of spatial-spectral information through a feature point matching algorithm.

[0090] Specifically, this step first relies on the synchronous collection mechanism of the visible light camera and the spectral imaging module. Both are synchronized by time stamp (such as IEEE 1588 precise time protocol) to ensure that the image collection time error is less than 1 ms, the spatial view angle is consistent (within ±0.5° deviation) to ensure the accuracy of subsequent alignment. Then, SIFT (Scale-Invariant Feature Transform) or improved SURF, ORB, etc. feature point extraction algorithm is used to extract key feature points with scale invariance and rotation invariance from the RGB image. These feature points usually include corner points, edge points, and other regions with significant texture changes, and their descriptors have a dimension of 128 (SIFT) or 64 (SURF), with good robustness and matching accuracy.

[0091] The feature point matching algorithm needs to meet the following requirements: the matching accuracy (re-projection error) should be less than 1 pixel, and the matching success rate (correct matching point ratio) should be higher than 85%. In addition, to deal with the possible view angle deviation and spatial distortion of the unmanned aerial vehicle platform during low-altitude flight, the system introduces the RANSAC (Random Sample Consensus) algorithm for error matching elimination and geometric transformation estimation to ensure that the finally aligned image has sub-pixel level spatial consistency. The matched transformation matrix (such as 3x3 affine matrix or 3D rigid transformation matrix) will be used for spatial registration of the spectral image, so that it is aligned with the RGB image at the pixel level.

[0092] This step is widely applicable to dynamic target identification and environmental monitoring in low-altitude remote sensing. For example, in agricultural remote sensing, visible light images can provide high-resolution crop distribution information, while spectral images can provide sub-angstrom-level spectral features of chlorophyll, water, pests, and diseases. Through cross-modal alignment, the system can accurately map spectral features to the corresponding area of the visible light image, realizing real-time and accurate assessment of crop health status. In disaster monitoring, this step can assist in identifying pollutant diffusion paths and improving emergency response efficiency.

[0093] This step effectively solves the problem of spectral positioning deviation in traditional spectral remote sensing systems due to the lack of spatial guidance. Through feature point matching, the system realizes pixel-level alignment of spectral data and visible light images, enabling spectral information to accurately correspond to the spatial location of features, thereby significantly improving identification accuracy and data fusion efficiency. This technology combines the advantages of computer vision and spectral imaging, providing an innovative path for cross-modal collaborative processing of low-altitude remote sensing systems.

[0094] Further, S4 comprises: S41, Gaussian-Laplacian operator preprocessing is performed on the visible light image to extract multi-scale edge feature points as a matching reference.

[0095] Specifically, this step aims to extract edge feature points with multi-scale characteristics from the visible light image as a reference for subsequent spatial alignment and feature matching with the hyperspectral image.

[0096] The LoG operator can effectively detect edge regions with significant gradient changes in the image by performing Gaussian smoothing and then Laplacian sharpening, while having multi-scale analysis capability, suitable for texture and edge feature extraction of different scales.

[0097] In specific operation, a multi-scale LoG filtering strategy is adopted to process the visible light image layer by layer. Usually, multiple scale parameters are set, and a corresponding LoG kernel (such as 5x5 or 7x7 size) is used for convolution operation at each scale. After convolution, edge points are identified by zero-crossing detection, i.e. finding points in the image where the LoG response value changes from positive to negative or from negative to positive, which correspond to the edge positions in the image. Further, the system performs non-maximum suppression and threshold screening on the edge points at multiple scales to remove noise responses and retain edge points with significant structural features.

[0098] The scale parameter of the LoG operator directly affects the sensitivity and scale adaptability of edge detection. Smaller values (such as 1.0) can detect small edges, suitable for vegetation texture or feature boundary identification; larger values (such as 2.5) are more sensitive to macroscopic structures (such as building outlines). In addition, the size of the LoG kernel is usually proportional to to ensure sufficient spatial response range. The LoG kernel size set in the system is , where is a constant (usually 2-3) to ensure good edge response at different scales.

[0099] This step is mainly used for cross-modal alignment between visible light images and hyperspectral images. Since the hyperspectral image usually has low spatial resolution, while the visible light image has high resolution and intuitive visual information, the multi-scale edge feature points extracted from the visible light image by the LoG operator can provide a stable and repeatable matching reference for subsequent SIFT and other feature matching algorithms, thereby achieving accurate registration of the hyperspectral image and the visible light image in the spatial dimension.

[0100] The multi-scale edge detection capability of the LoG operator significantly improves the robustness and adaptability of feature points, especially in complex feature scenarios (such as farmland, mining area, urban building group), which can stably extract key edge structures. This step provides a high-quality geometric alignment basis for subsequent AI-driven spectral-image fusion, effectively improving the spatio-temporal consistency and recognition accuracy of remote sensing data, and is an important supporting link for realizing real-time and high-precision feature recognition of sub-angstrom spectral remote sensing systems.

[0101] In S42, an improved SIFT algorithm is used to generate feature point descriptors in the spectral image, and a bidirectional nearest neighbor matching strategy is used to eliminate false matches caused by viewing angle differences.

[0102] Specifically, in some implementations, the improved SIFT (Scale-Invariant Feature Transform) algorithm extends the Gaussian difference pyramid structure of the traditional SIFT by introducing spectral dimension information, constructing a multi-band Gaussian difference space. The DOG (Difference of Gaussians) response is calculated for each pixel in multiple spectral bands, and stable feature points with significant spectral differences are extracted. Further, the feature point descriptor is not only based on the spatial gradient histogram, but also integrates the gradient direction and intensity information of multiple spectral channels to form a multi-dimensional feature vector, usually 128 dimensions or higher, to enhance the representation ability of the spectral characteristics of the features.

[0103] The improved SIFT algorithm sets the following key parameters in the spectral image: the number of scale space layers is 4, and the standard deviation σ of the Gaussian kernel increases between 0.5 and 1.6 at each layer; the feature point detection threshold is set to 0.03 to ensure that only significant feature points are retained; the descriptor window size is 16x16 pixels, the direction histogram is divided into 8 directions, and each direction contains gradient statistical information of multiple spectral channels. In the matching stage, the Mutual Nearest Neighbor (MNN) strategy is used, which calculates the nearest neighbor matching points in the visible light image and the spectral image for each feature point, and verifies that they are the nearest neighbors of each other, to eliminate false matches caused by viewing angle differences, lighting changes or geometric deformation. The matching threshold is usually set to 0.7, that is, if the matching distance ratio of two feature points is less than this value, the match is considered valid.

[0104] This step is mainly used to accurately align the visible light image and the hyperspectral image in the spatial dimension. Due to the attitude changes, lens distortion, and field of view differences between the spectral imaging module and the visible light camera during low-altitude flight of the unmanned aerial vehicle platform, the traditional one-way matching strategy is prone to introduce false matches. By combining the improved SIFT and MNN strategies, the robustness and accuracy of feature point matching can be effectively improved, providing a reliable spatial alignment basis for subsequent cross-modal data fusion and AI decision-making.

[0105] Also included are: S5, dynamically adjusting the encoding parameters of the resonant cavity array according to the spectral complexity of the ground object, calculating the encoding redundancy of each pixel point through a preset spectral complexity evaluation model.

[0106] Specifically, in some implementations, the present application dynamically adjusts the encoding parameters of the resonant cavity array through a preset spectral complexity evaluation model, thereby optimizing the collection efficiency and data quality of spectral information. The core of this step is to calculate and assign the corresponding encoding redundancy in real time according to the spectral feature complexity of the ground object at different pixel points, in order to achieve adaptive compression and information preservation of hyperspectral data.

[0107] This process first relies on the original spectral encoding image collected by the front-end optical system. The spectral information of each pixel point is modulated by multiple independently designed resonant cavity filter units, forming a multi-dimensional spectral response. In the back-end processing, the AI inference unit integrated on the chip quantifies the spectral complexity of each pixel point based on a preset spectral complexity evaluation model. This model can use a deep learning-based feature extraction network (such as CNN or Transformer architecture), with the resonant cavity response function and the light intensity distribution of the pixel point as input, and the spectral complexity score of the pixel point as output.

[0108] The input dimensions of the spectral complexity evaluation model include but are not limited to: spectral bandwidth (0.1 nm), slope of resonant cavity transmittance curve (> 10 dB / nm), entropy value of light intensity distribution (> 2.5 bits), and spectral similarity of adjacent pixels (< 0.75 correlation coefficient). The adjustment range of encoding redundancy is usually between 0.1 and 1.0, corresponding to the encoding density of different complexity regions, for example, setting the redundancy to 0.8 in the vegetation area (high spectral complexity), and setting it to 0.3 in the building area (low spectral complexity), to reduce the redundancy of invalid data.

[0109] This step realizes the intelligentization and adaptability of spectral information collection, effectively solving the problem of information loss or redundancy in complex ground object scenes with traditional fixed encoding strategy. By dynamically optimizing the encoding parameters, the system significantly improves the data collection efficiency and storage bandwidth utilization while ensuring sub-angstrom resolution, providing a high-quality, low-redundancy data foundation for subsequent real-time interpretation and multi-modal fusion.

[0110] S6, generating an adaptive compression encoding strategy based on the calculated redundancy, using high-density encoding for high-complexity regions and sparse encoding for low-complexity regions, reducing the data volume by more than 40%.

[0111] Specifically, this step is based on the quantitative evaluation of the spectral complexity of each pixel point in the spectral data cube, and local spectral entropy (Spectral Entropy, SE) and inter-band correlation (Inter-band Correlation, IBC) are used as complexity criteria. Specifically, for each pixel point, the spectral entropy value of the target band range (such as 350-1000nm, step 0.1nm) is calculated, and the higher the entropy value, the more complex the spectral information, and high-density coding is required; otherwise, the low-entropy area (such as the vicinity of the vegetation reflection peak) can use sparse coding strategy. In addition, combined with the spectral similarity analysis of adjacent pixels, the spatial consistency of the coding strategy is further optimized to avoid image edge distortion caused by local coding density mutation.

[0112] This strategy supports dynamic adjustment of coding redundancy, with a redundancy range of 1:1 to 1:10, and the coding density can vary between 100% and 30% according to the complexity of the region. For high-complexity areas (SE>0.8), a 1:1 redundancy is used to retain all band information; for medium-complexity areas (0.5<SE ≤ 0.8), a 1:4 redundancy is used to retain key bands; for low-complexity areas (SE ≤ 0.5), a 1:10 redundancy is used to retain only the main band information. This strategy is implemented on the FPGA side, with a processing delay controlled within 5ms, meeting the real-time requirements of low-altitude remote sensing of unmanned aerial vehicles.

[0113] This strategy is particularly suitable for scenarios that require high-precision spectral recognition, such as agricultural monitoring and environmental assessment. For example, in farmland, the spectral features of crop health status are usually concentrated in specific bands (such as 600-700nm), while the background soil or road spectrum changes less and can be sparsely coded, thereby reducing the overall data volume by more than 40% while ensuring recognition accuracy.

[0114] Through the adaptive compression strategy, the storage and transmission pressure of high-dimensional spectral data is effectively alleviated, the data processing efficiency of the system under limited bandwidth is improved, and the integrity of the key spectral information is ensured, providing a high-quality data foundation for subsequent AI inference and cross-modal fusion, significantly enhancing the practicality and adaptability of the low-altitude remote sensing system.

[0115] The method of the embodiment of the application further optimizes the spectral data acquisition efficiency by dynamically adjusting the coding parameters of the resonant cavity array and using an adaptive compression coding strategy, realizes differentiated coding processing for different ground object spectral complexities while ensuring sub-angstrom level ultra-high spectral resolution and high spatial resolution, effectively reduces the system data output by more than 40%, and improves the subsequent data processing and real-time analysis performance.

[0116] In summary, the present application uses the non-correlation of the spectral transmittance of each pixel to construct a unique high-dimensional spectral coding model. Traditional methods are mostly based on pixel correlation, which can easily lose information in complex scenes. The present chip can obtain multi-dimensional and multi-scale spectral details at the same time, such as distinguishing the subtle reflection differences of vegetation in different wavebands, providing a rich data basis for high-precision feature classification and recognition. Combined with the characteristics of heterogeneous spectral coding masks, an adaptive compression algorithm is designed. According to the complexity and importance of different pixel spectral information, the coding parameters are dynamically adjusted. In urban areas, the spectral data of key features such as buildings and roads are finely coded; in natural areas, vegetation and water bodies are compressed according to their spectral characteristics, maximizing compression efficiency and reducing data volume while ensuring the lossless of important information. An intelligent interpretation module is integrated on the CMOS chip, which realizes real-time interpretation based on pixel non-correlated spectral data. Using deep learning algorithms, the collected spectra are quickly analyzed, and the results such as feature types and material compositions are directly output. In agricultural low-altitude remote sensing, crop diseases, pests, and nutritional status can be identified in real time, providing a quick decision basis for precision agriculture. By fusing spectral remote sensing with other sensing data and expanding the fusion dimension with the pixel non-correlation characteristics, visible light images can be fused to supplement the image detail information with spectral data, and image information can assist spectral classification to improve the understanding of complex scenes, and the ecological status can be more comprehensively evaluated in environmental monitoring.

[0117] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, the person skilled in the art can combine and combine the different embodiments or examples described in the present specification and the features of the different embodiments or examples without contradiction.

[0118] In addition, the terms "first", "second" are only for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "a plurality of" is at least two, for example, two, three, etc., unless otherwise specifically limited.

Claims

1. A sub-angstrom resolution low-altitude spectral remote sensing system, characterized in that, The system comprises an optical imaging module, a spectrum acquisition module, a storage communication module and a comprehensive control module. The optical imaging module is used for capturing spatial image information of a target area and pre-processing incident light to provide high-quality incident light to the spectrum acquisition module, and transmitting the split image or independently acquired real-time image to the rear system to provide visual guidance for remote control of the remote sensing platform. The spectrum acquisition module is used for decomposing the incident light from the optical imaging module into a spectrum with sub-angstrom resolution and converting the spectrum into a corresponding electrical signal. The storage communication module is used for storing spectrum data represented by the electrical signal output by the spectrum acquisition module in real time, and performing bidirectional data interaction with the remote sensing platform, including flight parameters, system pose information, ground control instructions and data link state information. The comprehensive control module is used for coordinating the collaborative work of the optical imaging module, the spectrum acquisition module and the storage communication module, and performing environmental adaptive compensation control, real-time data processing and servo control.

2. An integrated pod device based on sub-Angstrom resolution low-altitude optical remote sensing system, characterized in that, The system comprises at least a central axis bundle, a base, an optical cabin and a comprehensive control board. The central axis bundle extends longitudinally, with its upper end connected to the unmanned aerial vehicle platform through the base and its lower end connected to the optical cabin. The base is internally integrated with an azimuth angle sensor and an azimuth motor. The azimuth angle sensor is used for collecting azimuth angle information of the base relative to the unmanned aerial vehicle platform in real time and transmitting the information to the comprehensive control board. The azimuth motor is used for receiving driving instructions from the comprehensive control board to drive the entire pod to rotate around the central axis bundle to adjust the overall orientation of the pod. The optical cabin is located at the lower part of the pod and is internally integrated with a sub-angstrom spectrum acquisition device, a visible light camera and an industrial computer. The sub-angstrom spectrum acquisition device is equipped with a sub-angstrom resolution low-altitude spectrum remote sensing chip and is used for collecting high-spectral resolution spectrum data of the target area. The optical axis of the visible light camera is parallel or coaxial with the viewing axis of the sub-angstrom spectrum acquisition device, and is used for synchronously collecting real-time visible light images of the target area to provide image transmission visual guidance for unmanned aerial vehicle remote control. The industrial computer is in communication connection with the unmanned aerial vehicle platform and the comprehensive control board, and is used for receiving control instructions, coordinating and controlling the working states of the sub-angstrom spectrum acquisition device and the visible light camera, and storing and forwarding the collected spectrum data and visible light images in real time. The integrated control board is integrated in the optical cabin, communicates with the flight control system and the image transmission system of the unmanned aerial vehicle platform through an open interface, receives flight parameters, control instructions and time synchronization signals from the unmanned aerial vehicle, and processes feedback signals of a pitch angle sensor, a pitch speed sensor and an azimuth speed sensor arranged in the optical cabin, and calculates the complete position information of the optical cabin in real time; based on the position information and the received control instructions, the integrated control board executes a servo control algorithm, generates a driving signal and outputs the driving signal to an azimuth motor and a pitch motor in the optical cabin, so as to realize stable pointing and dynamic tracking of the optical cabin in two degrees of freedom of azimuth and pitch, and complete stable imaging and remote sensing detection of a target area.

3. An encoding modulating chip for sub-Angstrom level spectral image snapshot acquisition, characterized in that, The chip comprises a silicon substrate, an array of tunable filter units integrated on the silicon substrate, and an AI inference processing unit. The array of tunable filter units is composed of a silicon nitride waveguide layer and an array of metal resonant cavities stacked on the silicon substrate, each metal resonant cavity serving as an independent filter unit, realizing a non-correlation designed spectral transmittance response by adjusting the physical structure, physically encoding and modulating the incident light, breaking the spectral correlation between pixels in traditional imaging, and synchronously capturing multi-dimensional spectral information with a resolution of 0.1 nm under single exposure. The silicon nitride waveguide layer is used for compressing and conducting incident light, enhancing the interaction between light and the metal resonant cavity, and improving the selective transmittance of the target waveband. The integrated AI inference processing unit is coupled to a CMOS or CCD sensor for receiving and decoding in real time the encoded light intensity signal output by the sensor after being modulated by the array of tunable filter units, and reconstructing the continuous fine spectral curve corresponding to each pixel point.

4. The chip of claim 3, wherein, The chip also supports cross-modal alignment function, which synchronously acquires visible light image and spectral image, and realizes feature point registration of multi-modal images by using SIFT or similar feature matching algorithm.

5. A sub-AE meter high-spectral resolution low-altitude remote sensing method, characterized in that, The chip comprises: S1, non-correlation encoding and modulation of incident light by the array of super surface resonant cavities, each resonant cavity independently designed to form a multi-dimensional spectral response without interference; S2, based on the light path compression structure of the integrated silicon nitride waveguide layer, the encoded spectral information is synchronously transmitted to the image sensor to generate an encoded light intensity map; S3, using the FPGA with AI inference unit to solve the encoded light intensity map in real time, and inversely reconstructing the spectral data cube with sub-angstrom resolution according to the transmission characteristics of each resonant cavity; S4, combined with the RGB image synchronously acquired by the visible light camera, the cross-modal alignment and fusion of spatial-spectral information is realized by the feature point matching algorithm.

6. The method of claim 5, wherein, The non-correlation encoding and modulation of incident light by the array of super surface resonant cavities, each resonant cavity independently designed to form a multi-dimensional spectral response, further comprises: S11, using a resonant cavity with a metal-insulator-metal structure to realize a transmission peak with a wavelength interval of 0.1 nm by adjusting the period, width and depth parameters of the resonant cavity; S12, by changing the refractive index distribution of the waveguide layer, the transmission rate of the resonant cavity is optimized by the coupling effect of the silicon nitride waveguide layer and the metal resonant cavity, so that the signal separation degree of adjacent wavebands is ≥95%.

7. The method of claim 5, wherein, The optical path compression structure based on the integrated silicon nitride waveguide layer synchronously transmits the encoded spectral information to the image sensor to generate an encoded light intensity map, and further comprises: S21, a three-dimensional optical path compression network is formed by a plurality of layers of silicon nitride waveguides, and the light propagation path is controlled by the refractive index difference between the waveguide layers, so that the spectral information forms a two-dimensional spatial distribution on the sensor plane; S22, a graded refractive index waveguide design is adopted, so that different wavebands of light produce a preset angle offset when propagating in the waveguide, realizing orthogonal separation of spectral dimension and spatial dimension.

8. The method of claim 5, wherein, The FPGA with an AI inference unit is used to perform real-time calculation on the encoded light intensity map, and the transmission characteristics of each resonant cavity are used to inversely calculate a spectral data cube with sub-angstrom resolution, and further comprises: S31, a convolutional neural network algorithm is executed by FPGA hardware acceleration to perform nonlinear inversion on the encoded light intensity value of each pixel point to generate a continuous spectrum curve; S32, an orthogonal matching pursuit algorithm is used to perform sparse decoding on the resonant cavity transmission matrix, and the calculation complexity is reduced by iterative optimization, so that the decoding delay is ≤5ms.

9. The method of claim 5, wherein, The RGB image synchronously collected by the visible light camera is combined to realize cross-modal alignment and fusion of spatial-spectral information through a feature point matching algorithm, and further comprises: S41, the visible light image is preprocessed by a Gaussian-Laplacian operator to extract multi-scale edge feature points as a matching reference; S42, an improved SIFT algorithm is used to generate feature point descriptors in the spectral image, and a bidirectional nearest neighbor matching strategy is used to eliminate mismatching caused by view angle difference.

10. The method of claim 5, wherein, Further comprising: S5, the encoding parameters of the resonant cavity array are dynamically adjusted according to the spectral complexity of the ground object, and the encoding redundancy of each pixel point is calculated by a preset spectral complexity evaluation model; S6, based on the calculation redundancy, an adaptive compression encoding strategy is generated, high-density encoding is used for high-complexity areas, and sparse encoding is used for low-complexity areas, so that the data volume is reduced by more than 40%.

Images