An astronomical telescope remote observation method and system based on artificial intelligence

Abstract

The application discloses an astronomical telescope remote observation method and system based on artificial intelligence, and comprises the following steps: through multi-device compatible access and a remote control unit, precise pointing, parameter configuration and auxiliary equipment control of a telescope are realized; through an AI intelligent observation auxiliary unit, real-time feedback images are subjected to automatic celestial body identification, intelligent exposure parameter recommendation, image noise reduction enhancement and automatic guide star correction, and based on environment monitoring, abnormal protection and target recapture are triggered; through a resource management unit, an observation archive and a structured knowledge base are constructed; and through an interactive unit, observation reservation, multi-terminal adaptation and science popularization interaction are provided. Through the above manner, the application can reduce the astronomical observation threshold, improve the imaging quality and observation success rate, and realize the collaborative sharing of observation resources and science popularization content.

Description

Technical Field

[0001] This invention relates to the field of astronomical observation and intelligent control technology, and in particular to a remote observation method and system for astronomical telescopes based on artificial intelligence. Background Technology

[0002] With the advancement of astronomical popularization and the expansion of the deep-sky photography enthusiast community, the public's demand for professional astronomical telescopes is increasing. However, existing technologies still face the following prominent problems in practical applications:

[0003] First, the equipment has a high barrier to entry. Professional astronomical telescopes are expensive and have complex operating procedures, involving professional knowledge such as equatorial mount calibration, polar alignment, and guide star parameter configuration. The general public and primary and secondary school teachers and students find it difficult to complete the observation operation independently.

[0004] Second, observation resources are unevenly distributed. High-quality astronomical observatories are usually located in remote areas far from urban light pollution, resulting in high equipment idle rates. Meanwhile, urban users lack access to these observatories and effective remote sharing and scheduling mechanisms.

[0005] Third, the level of intelligence is insufficient. Existing remote control solutions are mostly simple video transmission and mechanical remote control, lacking the ability to perceive and adaptively process environmental changes and celestial shifts in real time. When encountering cloud cover, sudden increases in wind force, or target misses, the system cannot automatically implement protection or recapture.

[0006] Fourth, there is a disconnect between observation and popular science. Current technologies focus on equipment control itself. After the observation is completed, the raw data obtained by the user lacks structured knowledge connections and cannot be effectively transformed into learnable and disseminated popular science resources, making it difficult to form a closed-loop ecosystem of "observation as learning".

[0007] Therefore, there is an urgent need for an intelligent observation solution for astronomical telescopes that can lower the professional threshold, improve the imaging success rate, achieve resource sharing, and integrate popular science functions. Summary of the Invention

[0008] The main technical problem solved by this invention is to provide a remote observation method and system for astronomical telescopes based on artificial intelligence. The aim is to reduce the operational threshold of astronomical observation, improve the success rate of observation and imaging quality, and realize the remote sharing of equipment resources and the systematic accumulation of popular science content.

[0009] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0010] An artificial intelligence-based method for remote observation with astronomical telescopes includes the following steps:

[0011] Step 1: During the observation mission, AI automatically identifies celestial objects in the images transmitted back by the telescope in real time, and intelligently recommends exposure parameters based on the identified celestial object type and real-time environmental data; the AI ​​algorithm is used to denoise and enhance the observed images, and the telescope is automatically adjusted after the guide mirror captures star points and AI detects star point shifts to achieve automatic guide correction.

[0012] Step 2: Monitor meteorological data and equipment status in real time. When abnormal conditions are detected, automatically trigger equipment protection actions and automatically calculate coordinates to control the telescope to retrieve the target when the celestial body leaves the field of view.

[0013] Step 3: Associate the observation data with celestial information to generate a structured observation archive, which is then stored as a popular science resource library. The popular science resource library includes celestial encyclopedia knowledge, graded observation tutorials, and astronomical event calendars.

[0014] Step 4 provides multi-user observation reservation, multi-terminal interaction, and science popularization interaction functions to form a closed loop of collaborative sharing of observation resources and science popularization content.

[0015] An artificial intelligence-based remote observation system for astronomical telescopes includes:

[0016] Application interaction layer: As the user side, it provides multiple entry points and various interaction methods;

[0017] Cloud service layer: As the cloud, it consists of three parts: AI computing center, business scheduling center and data center; the application interaction layer has multiple entry points and bidirectionally transmits HTTPS / WebSocket protocol commands with the business scheduling center, and each interaction method receives RTMP / HLS video streams / results transmitted from the data center;

[0018] Edge intelligence layer: As the edge, it includes core hardware nodes and software service nodes; the AI ​​computing center of the cloud service layer transmits sub-second level correction instructions to the core hardware nodes, the business scheduling center sends JSON control instructions to the core hardware nodes, and the AI ​​computing center receives the initial recognition effect of the compressed video stream sent by the software service nodes.

[0019] Sensing and Execution Layer: As the edge and field, it includes optical sensing systems, environmental sensing systems, and mechanical actuators, and connects to the core hardware nodes of the edge intelligence layer via UART / USB.

[0020] The beneficial effects of this invention are:

[0021] (1) Significantly lowers the threshold for observation. By using the YOLO algorithm to achieve automatic celestial object recognition, intelligent recommendation of exposure parameters, and one-click observation mode, novice users can obtain clear celestial object images without having professional astronomical knowledge, which greatly reduces the complexity of operation.

[0022] (2) Effectively improve imaging accuracy and observation success rate. By using convolutional neural network noise reduction, multi-frame superposition and automatic guiding correction technology, the interference of atmospheric jitter and sensor noise on long exposure imaging is overcome; combined with the automatic recapture mechanism for anomalies and intelligent retrieval after target loss, the observation success rate is significantly improved.

[0023] (3) A closed-loop ecosystem of observation and popular science has been constructed. The system automatically associates each observation data with astronomical encyclopedia information to generate structured observation archives, so that each observation result is transformed into a learnable and shareable popular science resource, solving the problems of data silos and popular science gaps in traditional astronomical observation.

[0024] (4) It has enabled the universal sharing of high-quality observation resources. Through mechanisms such as multi-user reservation scheduling and private equipment access sharing, professional telescope resources distributed in remote areas are efficiently connected with urban schools and science popularization institutions, which has improved the utilization rate of equipment and reduced the economic and time costs for the public to access professional astronomical observation. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of a four-layer simplified architecture of an artificial intelligence-based remote observation system for astronomical telescopes, according to a specific embodiment of the present invention.

[0026] Figure 2 This is a schematic diagram of a five-layer complete architecture of an artificial intelligence-based remote astronomical telescope observation system according to a specific embodiment of the present invention, compared to... Figure 1 It fully demonstrates the end-to-end logic of high concurrency, low latency, and AI collaboration.

[0027] Figure 1 and Figure 2 These are two hierarchical presentation methods for the system, not two independent implementation schemes. If necessary, they can be arranged according to... Figure 2 Let's elaborate.

[0028] Figure 3 This is a diagram of the AI-powered intelligent observation and guiding technology in a specific embodiment of the present invention.

[0029] Figure 4 This is a hardware topology and security closed-loop control diagram in a specific embodiment of the present invention.

[0030] Figure 5 This is a resource sharing and scheduling diagram in a specific embodiment of the present invention.

[0031] Figure 6 This is a diagram of the multi-segment collaboration and dual-mode interaction architecture in a specific embodiment of the present invention.

[0032] Figure 7This is a data flow and storage diagram in a specific embodiment of the present invention. Detailed Implementation

[0033] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings.

[0034] This embodiment consists of two main parts: a hardware execution terminal deployed at the observatory observation site and a user-facing web front-end interactive interface. The overall architecture is as follows: Figure 1 and Figure 2 As shown.

[0035] Reference Figure 1 Artificial intelligence-based remote astronomical telescope observation systems connect and communicate according to the following hierarchical relationship:

[0036] Application Interaction Layer: From the user's perspective, this layer provides multiple entry points via Web, App, and Mini Program, offering various interaction methods including dual-mode control, voice interaction, and science popularization interaction. The Web interface provides a full-featured console supporting advanced parameter configuration, including: telescope shooting parameters (exposure time, ISO, gain, focal length), celestial object position coordinates (right ascension RA, declination DEC), equatorial mount motion parameters, filter switching parameters, and guide star correction parameters. The App supports virtual joysticks, real-time preview, and voice control. The Mini Program provides a lightweight entry point, supporting reservations, viewing science information, and receiving reminders. Furthermore, dual-mode control offers a one-click observation mode and a professional business-grade mode; voice interaction enables voice command input and cloud-based ASR parsing; and science popularization interaction provides real-time bullet comments, celestial encyclopedia pop-ups, and the ability to share observation results.

[0037] Cloud Service Layer: Serving as the cloud and the brain, this layer comprises three parts: an AI computing center, a business scheduling center, and a data center. The application interaction layer uses multiple entry points to bidirectionally transmit HTTPS / WebSocket protocol commands to the business scheduling center, while various interaction methods receive RTMP / HLS video streams / results transmitted from the data center.

[0038] The AI ​​computing center includes the YOLOv8-Large high-precision model, which includes, but is not limited to, the following: Architecture: The backbone uses CSPDarknet53, the neck uses PAN-FPN with enhanced multi-scale fusion, and the detection head is a decoupled head, specifically optimized for faint nebulae, galaxies, and small objects in deep space; Training method: It is first pre-trained on the COCO dataset, and then fine-tuned using 200,000 astronomical survey images + system-measured deep-space images, supporting multi-category (star / planet / nebula / galaxy / star cluster) detection, deployed on cloud GPUs, with mAP ≥ 0.92, used for secondary high-precision verification of edge results, and secondary high-precision verification of image and video streams uploaded at the edge (identifying faint nebulae and galaxies); Star map matching algorithm: Based on the GAIA / UCAC4 star catalog, blind solution (Plate Solving) is performed to calibrate pointing errors; AI automatic guiding algorithm: It receives guiding data, calculates the centroid drift, and generates sub-arcsecond level correction instructions; Image enhancement algorithm: It uses a CNN denoising network and a multi-frame overlay algorithm to improve the signal-to-noise ratio.

[0039] The business scheduling center includes a user management system that implements real-name authentication and hierarchical permissions; a time-sharing reservation / queue engine that provides a Redis-based task queue and adopts a "one person operates, multiple people watch" concurrency logic; and device status monitoring that performs global heartbeat detection and manages the online / offline status of devices.

[0040] The data center includes an observation archive that stores structured images, environmental data, and results; a science popularization resource library containing astronomical encyclopedias, observation tutorials, and astronomical event calendars; and a teaching courseware library that supports K-12 courses and class-based observation records.

[0041] Edge Intelligence Layer: Serving as the edge and core hub, it includes core hardware nodes and software service nodes. The AI ​​computing center in the cloud service layer transmits sub-second-level correction instructions to the core hardware nodes, the business scheduling center sends JSON control instructions to the core hardware nodes, and the AI ​​computing center receives the initial recognition results of the compressed video stream sent by the software service nodes.

[0042] In the core hardware nodes, a circuit board is designed as the core control board and field master controller, replacing the Raspberry Pi / ESP32 combination. The circuit board receives coordinate commands and transmits them, while simultaneously running a lightweight YOLO deep learning model to perform real-time celestial object recognition and image analysis on the video stream returned by the camera.

[0043] The software service nodes include lightweight AI (YOLOv8-Tiny), which uses the INT8 quantization model to achieve real-time inference and rapid identification of bright celestial objects on the edge (>25FPS); communication services, which implement USB-to-serial port driver, use UART / I2C protocol stack and WiFi / 4G network pass-through; video encoding services, which use H.265 hardware encoding, dynamic bitrate adjustment (0.8-4Mbps), and push streaming to the cloud; and an anomaly safety protection mechanism, which monitors meteorological thresholds in real time and realizes anomaly detection operations of anomaly → interruption of exposure → shutdown → instrument return.

[0044] Sensing and Execution Layer: As the edge and field, it includes optical sensing systems, environmental sensing systems, and mechanical actuators, and connects to the core hardware nodes of the edge intelligence layer via UART / USB.

[0045] The optical sensing system includes a primary mirror camera, which uses an electronic eyepiece to capture high-definition video streams of the starry sky (1080P / 4K); and a guide mirror camera, which uses an independent optical path to capture guiding images for AI centroid calculation, i.e., the "improved weighted grayscale centroid method" mentioned earlier. In the automatic guiding algorithm, this calculation result is used to construct closed-loop feedback control. Specifically, the system compares the calculated star centroid coordinates with the preset "guide star" reference coordinates to obtain pixel-level offsets of the right ascension (RA) and declination (DEC) axes; then, based on a PID control algorithm, this pixel offset is converted into pulse correction commands for the stepper motor in real time, driving the equatorial mount to perform sub-arcsecond attitude fine-tuning, thereby offsetting mechanical errors and atmospheric disturbances, ensuring that the stars are rounded and free of streaks under long exposures.

[0046] The environmental sensing system includes a meteorological sensor group, comprising a wind speed sensor, a raindrop sensor, a light pollution detector, and a visibility meter; and equipment status sensors, which monitor voltage, temperature, and provide physical limit protection (limit switches) for mechanical movement limits. These limit switches define the physical boundaries of the equatorial mount's rotation, preventing the equipment from rotating indefinitely due to abnormal control signals and breaking cables, thus forming the last line of defense at the hardware level.

[0047] The mechanical actuators include the Onstep equatorial mount, which uses a dual-axis stepper motor (right ascension / declination) to support high-precision micro-stepping drive; an electric focuser, which uses a stepper motor to control the focal length; an electric filter wheel, which supports switching between UHC, H-alpha and other filters; and an automatic lens cap, which uses relay control to protect the lens.

[0048] One specific mechanical actuator involves deploying a personal computer next to the telescope as a control terminal, which performs two functions: first, running backend service programs to process observation requests sent from the user's front end; second, acting as a serial communication host to send and receive data with the circuit board. The control terminal maintains a communication connection with the user's front-end web interface device via a wired or wireless network.

[0049] When a user initiates an observation request via the front-end web interface, the system retrieves and calculates the precise right ascension (RA) and declination (DEC) coordinates of the celestial body at the current moment from the data center based on the user's input celestial body name. The edge computing circuit board generates a control command frame conforming to the OnStep communication protocol based on the received right ascension and declination coordinates and sends it to the OnStep equatorial mount control box via another serial communication interface (such as UART or I2C bus). The OnStep equatorial mount integrates a microcontroller and a dual-axis stepper motor drive circuit. Upon receiving the command, the control box outputs pulse signals of specific frequency and quantity to the right ascension and declination stepper motors, respectively. The two stepper motors drive the right ascension and declination axes of the equatorial mount to rotate, thereby causing the telescope's primary mirror and imaging camera to precisely point at the target celestial body. Actual testing shows that the pointing error of this system can be stably controlled within 0.1°.

[0050] Please note the following in the provided image: Figure 1 and Figure 2 It is not about repeatedly demonstrating the system, but about explaining the system at different depths and levels. Figure 1 It demonstrates the system's four-layer logical architecture, focusing on the abstract layering of business processes; and Figure 2 This demonstrates a 5-layer physical architecture, focusing on the specific hardware deployment in high-concurrency scenarios. Figure 2 Not an independent solution, but Figure 1 The specific mapping on the 'cloud-edge-device' physical devices—especially the physical entities that break down the 'edge intelligence layer' into 'core hardware nodes' and 'software service nodes'—clearly defines the physical boundaries and communication links between the circuit board, server cluster, and cloud center.

[0051] This embodiment also proposes a remote observation method for astronomical telescopes based on artificial intelligence, referring to... Figure 3The diagram illustrates the AI-powered end-to-end intelligent observation and guiding technology. "Target locking / ROI cropping" refers to a dynamic region of interest extraction technique based on motion vector prediction. Instead of brute-force calculations across the entire image, the system dynamically crops out the core region (ROI) containing the target celestial object based on a celestial motion model, and performs high-precision centroid tracking within this region. This significantly reduces computational power consumption at the edge, increasing the guiding frequency to over 25 FPS. Specifically, once the telescope is pointed at the target celestial object, the real-time video streams from the primary mirror camera and the guiding mirror camera are transmitted back to the circuit board via UART / USB interface, and the AI-assisted observation process is immediately initiated, including the following steps:

[0052] (a) YOLO Celestial Recognition

[0053] A lightweight YOLOv8n target detection model, specifically trained for astronomical images, is employed and customized for the characteristics of faint, small targets and low signal-to-noise ratio in astronomical images. The backbone network uses a CSPDarknet structure with a newly added small target enhancement branch to strengthen the capture of small-sized celestial features. The neck network uses a PAN-FPN structure with a CBAM attention mechanism to balance the extraction of features from bright and dark celestial objects. The detection head employs a decoupled head structure, setting five detection labels: stars, planets, nebulae, galaxies, and star clusters, outputting celestial object bounding boxes and confidence scores. The model is trained using transfer learning, first pre-trained on the COCO dataset, then fine-tuned and fully trained using a dedicated dataset containing 120,000 publicly available survey images and 80,000 images measured by this system. After INT8 quantization, it is adapted for edge deployment, achieving a single-frame detection time of ≤40ms and an average celestial object recognition accuracy of ≥95%, stably enabling real-time celestial object recognition. This model can analyze video frames in real time, quickly identify various celestial bodies (such as stars, planets, nebulae, and galaxies) appearing in the frame, and output celestial body type labels and their bounding box coordinates in the frame. The model's recognition results, along with confidence information, are sent back for subsequent processing and front-end display.

[0054] (II) Recommended Intelligent Exposure Parameters

[0055] Based on the celestial body type output by the YOLO model (e.g., identified as "M42 Orion Nebula" or "Jupiter"), combined with measured data from environmental perception hardware (light pollution detector, visibility meter) and AI visual analysis results from the electronic eyepiece image (sky background brightness, star point dispersion), multimodal environmental feature fusion is performed to determine the current light pollution level and atmospheric seeing. The circuit board then calls a preset exposure strategy algorithm library to automatically calculate and recommend a set of optimal shooting parameters, including exposure time, analog gain (ISO), and bias voltage. The automatic calculation of optimal shooting parameters is implemented using a three-order logic of "baseline mapping - environmental correction - threshold verification": First, a base parameter library of exposure time, ISO, and bias voltage corresponding to different celestial body types is established based on measured data; then, the base parameters are retrieved in combination with the identified celestial body type, and four correction coefficients for light pollution, seeing, celestial body elevation angle, and sky background brightness are introduced. The initial parameters are calculated through a nonlinear attenuation model based on empirical coefficients. The specific logic is as follows: using the baseline exposure value corresponding to the celestial body type as the base, light pollution level and seeing are introduced as negatively correlated correction factors for exponential attenuation correction, thereby dynamically generating an exposure time and ISO combination adapted to the current environment. Finally, a safety threshold check is performed on the initial parameters, limiting the maximum single-frame exposure to 300 seconds and the maximum ISO to 3200. The exposure rationality is verified by combining the image histogram. After the verification is passed, the settings are sent to the camera for configuration. Subsequently, the circuit board configures the above parameters to the main imaging camera via serial port commands, ensuring that appropriate exposure effects can be obtained for different celestial body types and observation environments, effectively avoiding overexposure or underexposure.

[0056] (III) Automatic Guiding Star Correction

[0057] During long exposures, due to factors such as mechanical transmission errors of the equatorial mount and changes in atmospheric refraction, the position of celestial objects in the image may slowly drift. The circuit board uses the YOLO algorithm to continuously monitor the centroid position of stars in the image returned by the guide mirror. The YOLO algorithm used here is the same lightweight YOLOv8n architecture and guide mirror-specific fine-tuned weight model as the celestial object recognition model mentioned above. It only retains the single-class detection label of "valid star points" and removes classification branches. The detection time per frame is ≤20ms, and edge computing power can be reused. The star centroid monitoring process is as follows: First, the guide mirror image is preprocessed for noise reduction. Then, 3-5 stable reference stars are detected and selected using this model. The sub-pixel level centroid coordinates of each reference star are calculated using an improved weighted gray-scale centroid method (introducing Gaussian point spread function fitting). This method is specifically optimized for the "tailing" or "diffuse" phenomenon of star points in astronomical images caused by atmospheric disturbances. Compared with the traditional geometric center method, the positioning accuracy is improved by more than 50%, and the average value is then used to obtain the reference centroid of the current frame. Using the centroid coordinates of the initial exposure frame as a reference, the offset is calculated frame by frame at a frequency of 10Hz to complete the real-time monitoring of the star point position. Once the star point offset is detected to exceed the preset threshold, the circuit board immediately calculates the offset vector. The circuit board establishes a pixel coordinate system with the initial reference centroid as the origin, calculates the horizontal and vertical pixel offsets of the current frame centroid, and obtains the pixel-level offset vector. Through the pre-calibrated camera pixel equivalent, the pixel offset is converted into the right ascension and declination arcsecond offsets in the equatorial coordinate system. Combined with the equatorial mount transmission ratio and motor step angle, it is converted into the number of reverse correction pulses of the stepper motor, generating fine-tuning control commands. The reverse correction amount is converted into pulse fine-tuning commands and sent to the OnStep control box in real time to drive the stepper motor to perform sub-arcsecond attitude correction. Through this closed-loop control mechanism, the system can stably control the star offset within 1 pixel during a single exposure that lasts for tens of minutes, ensuring that the stars in deep-space photography images are round and clear in detail.

[0058] (iv) Image noise reduction and enhancement processing

[0059] After the exposure, the circuit board performs preliminary noise reduction processing on the acquired raw image frames. Specifically, this includes: employing a Convolutional Neural Network (CNN) denoising model. This model is a lightweight CNN denoising model based on an improved U-Net architecture, specifically trained for astronomical images, and not a general-purpose existing model. A paired dataset of 60,000 pairs of noisy astronomical images measured by this system and corresponding standard noise-free images were constructed for training. After lightweight compression, it was adapted for edge-end operation, effectively eliminating sensor thermal noise and readout noise while preserving celestial details. After eliminating sensor thermal noise and readout noise, a multi-frame overlay algorithm is used to align and overlay multiple consecutively captured short-exposure images.

[0060] The specific steps of the multi-frame overlay algorithm are as follows: (1) Using the first frame of the sequence as the reference frame, complete the dark field and flat field correction for all images to be overlaid; (2) Extract the same star points of the reference frame and the frame to be overlaid, solve the homography transformation matrix through the RANSAC algorithm, and complete the image registration; (3) Use the Sigma cropping mean overlay method to remove abnormal gray values ​​pixel by pixel and take the mean to complete the image overlay; (4) Optimize the dynamic range of the overlaid image by gray-scale stretching and output a high signal-to-noise ratio image. The processed image data is uploaded for storage and further processing. The further processing here is based on the generation of popular science resources and the backtracking of abnormal states in the following text, and includes three supplementary processing: First, image standardization archiving, complete the format conversion, parameter extraction and structured data encapsulation; Second, optional advanced AI image optimization, including detail sharpening, color difference correction and light pollution stripping; Third, the observation process log association archiving, matching the equipment operation, guide star correction and environmental monitoring data of the corresponding time period for fault investigation and effect review.

[0061] This embodiment also provides an intelligent handling mechanism for abnormal situations, which monitors two types of abnormal states in real time during operation and takes corresponding protection and recovery measures:

[0062] (1) Environmental anomaly protection: refer to Figure 4 The circuit board connects to a wind speed sensor and a rain sensor via GPIO interfaces to continuously collect on-site meteorological data. It also receives real-time video streams of star drift and deep-sky images via USB. When a sudden increase in wind speed to level 5 or above is detected, or when the rain sensor detects rainfall, the system automatically determines this as an environmental anomaly and immediately executes a preset protection sequence: stopping the current exposure task, writing the captured raw data to local memory, sending a command via serial port to close the telescope tube cover, and driving the equatorial mount back to a safe position. These protective measures effectively prevent damage to the equipment due to severe weather.

[0063] (2) Target Loss Recapture: During observation, if a target celestial body leaves the camera's field of view due to brief cloud cover, bird flight, or signal interference, the YOLO model on the circuit board will detect the target loss event. Based on the last valid recorded celestial body coordinates and the star map matching algorithm, the precise pointing coordinates at the current moment are recalculated. This recalculation process adopts a three-order closed-loop process of "coordinate correction - pointing adjustment - YOLO verification," rather than simply repeating the YOLO celestial body recognition steps: First, the last valid reference coordinates of the target are retrieved, and combined with the current moment and station information, the deviations caused by the diurnal motion of the celestial body and atmospheric refraction are corrected, and the real-time apparent coordinates are recalculated; Second, the telescope is driven to complete the pointing adjustment based on the new coordinates; Third, the YOLO celestial body recognition model is called to verify whether the target has returned to the field of view. If there is a deviation, it is fine-tuned through the guiding process; if the target is not identified, the pointing is iteratively corrected through star map matching until the target is recaptured. The telescope is driven again to retrieve the target. According to actual tests, the target recapture success rate is higher than 98%.

[0064] This embodiment also designs a method for front-end webpage interaction and sharing of popular science resources, referring to... Figure 5 Users access this system through a front-end web interface:

[0065] (1) User and Business Scenarios: On the front-end interactive interface, users are presented with two scenarios: teaching scenarios and personal observation scenarios. They can initiate observation requests by entering the name of a celestial body, selecting it from the star catalog list, or directly clicking the "One-Click Observation" button. The teaching scenario is mainly for K12 / university classrooms, while personal observation is for users who make appointments or are on temporary assignments. For novice users, the system provides a "One-Click Observation" mode. Users only need to select the target, and the system will automatically complete the entire process of coordinate analysis, pointing control, parameter configuration, automatic shooting, and image optimization. For professional users, the advanced mode supports manually setting all parameters such as exposure time, gain, and filter wheel position, and supports writing custom observation sequences through scripting languages.

[0066] (2) Intelligent scheduling strategy: After receiving the observation request, the teaching time slice management is first implemented, dividing the night into N time slices for fragmented resource utilization. Then, a priority queue is established based on the priority arbitration rules of teaching tasks, reserved users and temporary tourists. Finally, based on meteorological environmental data and light pollution map, the best available site is recommended and fed back to the user to balance the overall load of the observation system.

[0067] (3) Heterogeneous equipment access: Based on the best available site recommended by the intelligent scheduling strategy, public observatories (public telescope clusters) or private equipment (P2P shared private telescopes) are allocated. Public observatories are fixed sites with a unified standard interface and high availability; private equipment are distributed sites that use SDK access or hardware modification interfaces to utilize idle observation resources.

[0068] Users with private astronomical telescopes can connect their equipment to the system by installing circuit board cards and the OnStep control box. Users can set the shareable time slots for their equipment in the backend, making it a "shared observation point." When other users book and use the equipment through the platform, the system automatically performs task scheduling and access management, further expanding the system's equipment sharing ecosystem.

[0069] Reference Figure 6 This embodiment also presents a multi-segment collaborative and dual-mode interaction method. In the front-end interaction layer, voice interaction functionality is implemented through multiple entry points, allowing switching between a one-click observation mode and an advanced professional mode. The one-click observation mode employs a fully automated AI process of automatic star acquisition → automatic focusing → automatic imaging, while the advanced professional mode provides customizable parameter functions, allowing users to adjust various parameters such as exposure time, gain, filters, and equatorial mount speed. In the back-end service layer, an intelligent scheduling engine, a science popularization engine, and a voice processing center are designed and deployed. The intelligent scheduling engine receives observation task requests and simultaneously parses the system instructions recognized and processed by the voice processing center, providing time-sharing reservation management and FIFO / priority-based queuing logic, and accepting access from public or private devices according to device sharing logic. The science popularization engine links to astronomical encyclopedias such as Wikipedia and professional databases, and recommends the best target based on the current weather / season / AI assessment.

[0070] Reference Figure 7 This embodiment presents a data flow and storage method. First, raw image data, shooting metadata, and environmental data are collected at the source. Then, data processing and storage are performed, employing an ETL pipeline for automated data cleaning and packaging, associating images with shooting parameters, removing invalid data, and performing format conversion. Subsequently, a structured archive is used to generate observation archives, employing SQL / NoSQL structured storage and generating unique index IDs. Finally, the stored data is applied, for example, stored in a teaching resource library as teaching material for smart education or sent to student terminals. For the needs of primary and secondary school astronomy teaching, a class-based group reservation function is supported. Course tasks such as "Lunar Crater Observation" or "Planetary Orbit Observation" can be created in the backend. The system queues and schedules requests from multiple students in the same class, controlling the telescope to point at the predetermined target in turn, and synchronously displaying the real-time images on each student's front-end web interface. Accompanying teaching materials (PPT, lesson plans) can be downloaded synchronously, realizing a complete teaching loop of "observation-recording-discussion". Additionally, an interactive science community can be established, with interest groups categorized by celestial body type (such as planetary observation, deep-sky photography, and meteor monitoring). Users can share their observations, exchange photography techniques, and initiate group observation invitations within these groups. The platform regularly hosts online observation challenges (such as the "Messier Astronomical Object Challenge Marathon") and astrophotography competitions to stimulate user participation and promote the dissemination of scientific knowledge.

[0071] In summary, this embodiment provides an intelligent, user-friendly, and shared astronomical observation solution by deeply integrating the YOLO deep learning algorithm with remote control hardware for astronomical telescopes and constructing a closed-loop management system for observation data and popular science resources. It effectively solves the core problems of high equipment threshold, insufficient intelligence, and gaps in popular science in existing technologies.

[0072] The above description is only a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. Any equivalent modifications or changes made by those skilled in the art based on the content disclosed in the present invention should be included within the scope of protection set forth in the claims.

Claims

1. An artificial intelligence-based astronomical telescope remote observation method, characterized in that: The method includes the following steps: Step 1: During the observation mission, AI automatically identifies celestial objects in the real-time images transmitted from the telescope. Based on the identified celestial object type and real-time environmental data, intelligent exposure parameters are recommended. The observed images are then denoised and enhanced using AI algorithms. A guide scope is used to capture star points, and AI detects star point shifts and automatically controls the telescope for fine-tuning, achieving automatic guide correction. The AI ​​automatic celestial object identification includes: using the real-time images transmitted from the telescope, AI automatically identifies the type of celestial object, labeling its class, distance, and apparent diameter; combining real-time meteorological data such as wind speed, visibility, light pollution level, and celestial object elevation angle, AI automatically recommends the best observation target for the day and generates suggested observation times. Celestial object identification uses a lightweight YOLOv8n target detection model specifically trained for astronomical images, with the backbone network using... The model employs a CSPDarknet architecture with a newly added small target enhancement branch to enhance the capture of features from small celestial objects. The neck region utilizes a PAN-FPN architecture with a CBAM attention mechanism to balance the extraction of features from bright and dark celestial objects. The detection head uses a decoupled head structure, setting five detection labels: stars, planets, nebulae, galaxies, and star clusters, and outputs celestial object bounding boxes and confidence scores. The model is trained using transfer learning, first pre-trained on the COCO dataset, then fine-tuned and fully trained using a dedicated dataset containing publicly available survey images and measured images, and finally adapted for edge deployment after INT8 quantization. The model analyzes video frames in real time, identifies various celestial objects appearing in the frame, and outputs celestial object type labels and their bounding box coordinates within the frame. The model's recognition results, along with confidence information, are fed back for subsequent processing and front-end display. Step 2: Monitor meteorological data and equipment status in real time. When abnormal conditions are detected, automatically trigger equipment protection actions and automatically calculate coordinates to control the telescope to retrieve the target when the celestial body leaves the field of view. Step 3: Associate the observation data with celestial information to generate a structured observation archive, which is then stored as a popular science resource library. The popular science resource library includes celestial encyclopedia knowledge, graded observation tutorials, and astronomical event calendars. Step 4 provides multi-user observation reservation, multi-terminal interaction, and science popularization interaction functions to form a closed loop of collaborative sharing of observation resources and science popularization content.

2. The artificial intelligence-based astronomical telescope remote observation method according to claim 1, characterized in that: The intelligent recommended exposure parameters include: automatically matching the optimal exposure time and ISO parameters according to the type of celestial body and the level of light pollution to avoid overexposure or underexposure; the noise reduction and enhancement processing includes: processing the observed image through CNN noise reduction and multi-frame overlay algorithms to remove atmospheric jitter and noise interference and improve image clarity.

3. The artificial intelligence-based astronomical telescope remote observation method according to claim 1, characterized in that: The abnormal conditions include: sudden rainfall, wind speed greater than or equal to level 5, telescope overload or power supply abnormality; the equipment protection actions include: closing the telescope lens cap, stopping exposure, and saving the captured data.

4. The artificial intelligence-based astronomical telescope remote observation method according to claim 1, characterized in that: The multi-user observation reservation includes: providing reservations for public telescopes by time slot, and allowing private telescopes to access and share resources, with settings for sharing time slots and pricing standards.

5. The artificial intelligence-based astronomical telescope remote observation method according to claim 1, characterized in that: The multi-terminal interaction includes: support for access and synchronized functions across multiple platforms such as Web, iOS / Android APP and mini-program; providing two operation interfaces: one-click observation mode and advanced mode. The one-click observation mode automatically matches parameters and takes pictures automatically, while the advanced mode supports full parameter customization and script control; the APP supports voice command control, including voice control for pointing to celestial bodies, starting exposure, and zooming in on the image.

6. An artificial intelligence-based astronomical telescope remote observation system, characterized in that: The system includes: Application interaction layer: As the user side, it provides multiple entry points and various interaction methods; Cloud service layer: As the cloud, it consists of three parts: AI computing center, business scheduling center and data center; the application interaction layer has multiple entry points and bidirectionally transmits HTTPS / WebSocket protocol commands with the business scheduling center, and each interaction method receives RTMP / HLS video streams / results transmitted from the data center; In the cloud service layer: The AI ​​computing center includes: a YOLOv8-Large model for secondary high-precision verification of image and video streams uploaded from the edge, identifying faint nebulae and galaxies; a star map matching algorithm based on the GAIA / UCAC4 star catalog for blind solution and calibration of pointing errors; an AI automatic guiding algorithm that receives guiding data, calculates the centroid drift, and generates sub-arcsecond correction commands; and an image enhancement algorithm that uses a CNN denoising network and multi-frame overlay algorithm to improve the signal-to-noise ratio. The business scheduling center includes a user management system that implements real-name authentication and hierarchical access control; a time-sharing reservation / queue engine that provides a Redis-based task queue and uses concurrent logic; and a device status monitoring system that performs global heartbeat detection and manages the online / offline status of devices. The data center includes an observation archive that stores structured images, environmental data, and results; a science popularization resource library containing astronomical encyclopedias, observation tutorials, and astronomical event calendars; and a teaching material library that supports K-12 courses and class observation records. Edge intelligence layer: As the edge, it includes core hardware nodes and software service nodes; the AI ​​computing center of the cloud service layer transmits sub-second level correction instructions to the core hardware nodes, the business scheduling center sends JSON control instructions to the core hardware nodes, and the AI ​​computing center receives the initial recognition effect of the compressed video stream sent by the software service nodes. Sensing and Execution Layer: As the edge and field, it includes optical sensing systems, environmental sensing systems, and mechanical actuators, and connects to the core hardware nodes of the edge intelligence layer via UART / USB.

7. The artificial intelligence-based remote observation system for astronomical telescopes according to claim 6, characterized in that: In the edge intelligence layer: In the core hardware nodes, the Hais board is used as the core control board and field master controller. It receives coordinate commands and transmits them. At the same time, it runs a lightweight YOLO deep learning model to perform real-time celestial object recognition and image analysis on the video stream returned by the camera. The software service nodes include lightweight AI, which uses an INT8 quantization model to enable real-time inference and rapid identification of bright celestial bodies on the edge; communication services, which implement USB-to-serial port drivers, use UART / I2C protocol stacks and WiFi / 4G network pass-through; video encoding services, which use H.265 hardware encoding, dynamic bitrate adjustment, and push streaming to the cloud; and image enhancement algorithms, which monitor meteorological thresholds in real time and implement anomaly detection operations: anomaly → interruption of exposure → shutdown → instrument return.

8. The artificial intelligence-based remote observation system for astronomical telescopes according to claim 7, characterized in that: In the perception and execution layer: The optical sensing system includes a primary mirror camera, which uses an electronic eyepiece to capture high-definition video streams of the starry sky; and a guide mirror camera, which uses an independent optical path to capture guiding images for AI centroid calculation. The environmental sensing system includes a meteorological sensor group, which includes a wind speed sensor, a raindrop sensor, a light pollution detector, and a visibility meter; and equipment status sensors, which realize voltage monitoring, temperature monitoring, and limit switches. The mechanical actuators include the Onstep equatorial mount, which uses a right ascension / declination dual-axis stepper motor to support high-precision micro-stepping drive; The electric focuser uses a stepper motor to control the focal length; the electric filter wheel supports filter switching; and the automatic lens cap uses a relay control to protect the lens.

Images