Intelligent telescope scenery image acquisition and processing system based on artificial intelligence

Abstract

The invention discloses an intelligent telescope scenery image acquisition and processing system based on artificial intelligence, which relates to the technical field of image communication and comprises a multi-telescope collaborative acquisition and fusion module, a distributed networking module, a content perception compression and transmission module, a feature encryption isolation module and a terminal-cloud collaborative communication module. According to the invention, through an integrated architecture design of multi-mirror collaborative acquisition fusion, distributed networking, content awareness compression transmission, feature encryption isolation and end-cloud collaborative communication, full-process collaborative optimization of scenery image acquisition, transmission, encryption and end-cloud interaction is realized. The integrity of image acquisition, the adaptability of transmission, the security of data and the efficiency of end-cloud collaboration can be remarkably improved, so that the problem that links of acquisition, transmission, encryption and end-cloud collaboration in an existing intelligent telescope image acquisition and processing system are split can be solved; therefore, the technical problems of poor image quality, low transmission efficiency, large data safety hidden trouble and poor terminal-cloud interaction adaptability are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image communication, in particular to an intelligent telescope scene image acquisition and processing system based on artificial intelligence. BACKGROUND

[0002] The intelligent telescope scene image acquisition and processing system provides accurate acquisition and optimization support for remote scene images of intelligent telescopes, captures clear original images under different environmental conditions, compensates for image quality loss in remote observation through intelligent processing, outputs high-quality visual information, and helps users efficiently complete remote scene observation, recording and analysis, meeting the remote visual needs in outdoor observation or scientific exploration scenes.

[0003] Currently, the intelligent telescope scene image acquisition and processing system has the problem of mutual fragmentation in the aspects of acquisition, transmission, encryption and end-to-cloud collaboration, and lacks effective fault redundancy mechanisms, differentiated resource allocation strategies and adaptive security and algorithm iteration schemes, resulting in poor image quality, low transmission efficiency, large data security risks, poor end-to-cloud interaction adaptability in actual application, and other problems such as data loss during communication failure, serious resource waste during transmission, heavy processing delay caused by heavy feature extraction burden on the end side, and mismatch between encryption methods and image sensitivity, model iteration easily leaking privacy and being disconnected from actual acquisition needs on the end side, which makes it difficult to meet the high-quality and efficient remote visual needs in outdoor observation or scientific exploration scenes.

[0004] Therefore, the intelligent telescope scene image acquisition and processing system based on artificial intelligence is proposed to solve the above problems. SUMMARY

[0005] The main purpose of the present application is to provide an intelligent telescope scene image acquisition and processing system based on artificial intelligence to solve the problems raised in the background.

[0006] To achieve the above purpose, the technical solution adopted by the present application is as follows: an intelligent telescope scene image acquisition and processing system based on artificial intelligence, which comprises a multi-lens cooperative acquisition and fusion module, a distributed networking module, a content-aware compression and transmission module, a feature encryption and isolation module and an end-to-cloud collaborative communication module. The multi-lens cooperative acquisition and fusion module assigns acquisition tasks to multiple intelligent telescopes, adapts the image feature data format and image communication protocol, and fuses to generate panoramic images or stereoscopic images. The distributed networking module realizes wireless interconnection of multiple intelligent telescopes, constructs a distributed acquisition communication network, and performs image communication fault redundancy processing. The content-aware compression transmission module identifies the importance of the scene image region, and completes the differentiated compression of the image data and the adjustment of the transmission parameters in combination with the state of the wireless communication link. The feature encryption isolation module extracts the abstract features of the scene image, identifies the sensitivity degree of the image, and processes the abstract features by using an encryption method matched with the sensitivity degree. The end-to-cloud collaborative communication module completes the image reconstruction, model iteration and image communication protocol switching required for image communication between the end and the cloud based on the encrypted abstract features.

[0007] Preferably, the multi-mirror collaborative acquisition fusion module includes an AI acquisition task scheduling unit, a multi-end data communication adaptation unit and a multi-source image feature fusion unit. The AI acquisition task scheduling unit allocates the image acquisition angle, the spectral acquisition channel and the acquisition time sequence to each intelligent telescope in the distributed acquisition communication network. The multi-end data communication adaptation unit unifies the image feature data format and the image communication protocol output by multiple intelligent telescopes. The multi-source image feature fusion unit fuses the image feature data output by multiple intelligent telescopes into a panoramic image or a stereoscopic image through feature matching and pixel-level alignment. When the multi-source image feature fusion unit fuses the image feature data, it is specifically used for: Receiving the image feature data output by multiple intelligent telescopes, the image feature data including the contour topology and texture association of the acquisition region of each device; Performing feature matching on all image feature data to establish a corresponding relationship between the image feature data of different devices; Performing pixel-level alignment based on the corresponding relationship to eliminate the spatial offset of the image feature data; Integrating the aligned image feature data to generate a panoramic image or a stereoscopic image.

[0008] Preferably, the distributed networking module includes a distributed networking management unit and a communication fault redundancy processing unit. The distributed networking management unit realizes the wireless interconnection of multiple intelligent telescopes and constructs a distributed acquisition communication network. The communication fault redundancy processing unit monitors the image communication state of each intelligent telescope in the distributed acquisition communication network, switches to a standby image communication link and schedules surrounding intelligent telescopes to complete image acquisition data when the image communication of any intelligent telescope is interrupted.

[0009] Preferably, when the communication fault redundancy processing unit processes the image communication fault, it is specifically used for: Continuously monitor the image communication connection state of each intelligent telescope, when detecting that the image communication of a certain intelligent telescope is interrupted, start the standby image communication link to attempt to reconnect; If the reconnection fails, determine the preset image acquisition area and acquisition task of the intelligent telescope; Query the idle intelligent telescopes in the communication coverage range of the failed equipment in the distributed acquisition communication network, and distribute the completed image acquisition task to the idle intelligent telescopes; Receive the image feature data after the idle intelligent telescopes complete the completed image acquisition task, which is used for subsequent image fusion.

[0010] Preferably, the content-aware compression transmission module includes an image content importance identification unit, a dynamic compression strategy generation unit, a communication link state monitoring unit, and a transmission parameter adaptive adjustment unit; The image content importance identification unit identifies the importance of different regions in the scene image; The dynamic compression strategy generation unit generates a differentiated compression strategy based on the region importance; The communication link state monitoring unit monitors the signal strength and stability of the wireless communication link; The transmission parameter adaptive adjustment unit adjusts the transmission frame rate of the image data packet based on the communication link state and the compression processing result.

[0011] Preferably, the feature encryption isolation module includes an end-side lightweight feature extraction unit and an end-side hierarchical encryption unit; The end-side lightweight feature extraction unit extracts abstract features of the scene image, and the abstract features include contour topology and texture association; The end-side hierarchical encryption unit identifies the scene type of the scene image, determines the image sensitivity, and processes the extracted abstract features using a hierarchical encryption method matched with the sensitivity.

[0012] Preferably, when the end-side lightweight feature extraction unit extracts the abstract features of the scene image, it is specifically used for: Performing pixel-level feature analysis on the scene image to separate contour topology-related features and texture association-related features; Structurally extracting the contour topology-related features to form contour topology, and performing association analysis on the texture association-related features to form texture association, and integrating the contour topology and the texture association to obtain the abstract features of the scene image.

[0013] Preferably, when the end-side hierarchical encryption unit identifies the image sensitivity and processes the abstract features, it is specifically used for: Performing scene semantic identification on the scene image to determine whether it contains human information or classified information, and determining the image sensitivity; An asymmetric encryption algorithm is used for high-sensitive images containing human information or classified identification, and a symmetric encryption algorithm is used for low-sensitive images not containing the above information, and the extracted abstract features are encrypted to generate encrypted abstract features.

[0014] Preferably, the end-cloud collaborative communication module comprises a cloud-end feature reconstruction optimization unit, a federated learning model iteration unit and a communication protocol dynamic adaptation unit. The cloud-end feature reconstruction optimization unit receives the encrypted abstract features and completes image reconstruction based on the abstract features. The federated learning model iteration unit receives model data uploaded by the end side, aggregates the model data to update the global image processing algorithm, and feeds back the updated global image processing algorithm to the end side. The communication protocol dynamic adaptation unit monitors the communication link state and switches the image communication protocol based on the communication link state.

[0015] Preferably, when the federated learning model iteration unit aggregates model data and updates the algorithm, it is specifically used for: Receiving model data uploaded by multiple intelligent telescope end sides respectively, the model data being associated with abstract features of scene images collected by each end side; Calculating the credibility weight of each end side model data, the formula being: In the formula, W i is the credibility weight of the model data of the i-th intelligent telescope end side, Q i is the collection reliability coefficient of the i-th intelligent telescope end side, F i is the abstract feature quality coefficient corresponding to the gradient data uploaded by the i-th intelligent telescope end side, n is the total number of intelligent telescope end sides participating in federated learning in the distributed collection communication network, Q j is the collection reliability coefficient of the j-th intelligent telescope end side, F j is the abstract feature quality coefficient corresponding to the gradient data uploaded by the j-th intelligent telescope end side. Weighted aggregation is performed on all model data based on the credibility weight to generate aggregated gradient data. Parameter updating is performed on the global image processing algorithm based on the aggregated gradient data to obtain an updated global image processing algorithm. The updated global image processing algorithm is fed back to each intelligent telescope end side.

[0016] The present application has the following advantages: ​1.The present application realizes the whole-process collaborative optimization of scene image acquisition, transmission, encryption and end-cloud interaction through the integrated architecture design of multi-mirror collaborative acquisition fusion, distributed networking, content-aware compression transmission, feature encryption isolation and end-cloud collaborative communication, which can significantly improve the integrity of image acquisition, the adaptability of transmission, the security of data and the efficiency of end-cloud collaboration compared with the prior art, thus solving the technical problems of poor image quality, low transmission efficiency, great data security risks and poor end-cloud interaction adaptability caused by the fragmentation of acquisition, transmission, encryption and end-cloud collaboration in the existing intelligent telescope image acquisition and processing system.

[0017] 2.The present application realizes fault-tolerant completion of acquisition data, precise allocation of transmission resources and efficient lightweight processing on the end side through the organic combination of distributed networking communication fault redundancy recovery mechanism, differential compression strategy based on image region importance and end-side lightweight abstract feature extraction technology, which can improve the reliability of image acquisition, the utilization rate of transmission bandwidth and the running efficiency of end-side devices compared with the prior art, thus solving the technical problems of data loss during communication failure, serious resource waste during transmission and heavy processing delay caused by heavy feature extraction burden on the end side in the existing system.

[0018] 3.The present application realizes the scene adaptability of data encryption and the dynamic optimization of image processing algorithm through the hierarchical encryption method based on image sensitivity and the end-cloud collaborative model iteration mechanism driven by federated learning, which can improve the pertinence of data encryption and the adaptability of algorithms to actual acquisition scenarios, while ensuring data privacy and security, thus solving the technical problems of mismatch between encryption method and image sensitivity, privacy leakage caused by model iteration relying on centralized data and disconnection between algorithm update and actual acquisition demand on the end side in the existing system. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 It is the overall system architecture diagram of the present application; Figure 2 It is the multi-mirror collaborative acquisition fusion module architecture diagram of the present application; Figure 3 It is the distributed networking module architecture diagram of the present application; Figure 4 It is the content-aware compression transmission module architecture diagram of the present application; Figure 5 It is the feature encryption isolation module architecture diagram of the present application; Figure 6 It is the end-cloud collaborative communication module architecture diagram of the present application. DETAILED DESCRIPTION

[0020] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments belong to some of the embodiments of the present application but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0021] The terms used in the embodiments of the present application are only for the purpose of describing particular embodiments and are not intended to limit the present application. The singular forms "said" and "this" used in the embodiments of the present application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. "Plural" generally includes at least two.

[0022] Depending on the context, the word "if" or "if" as used herein can be interpreted as "when" or "when" or "in response to determining" or "in response to detecting". Similarly, depending on the context, the phrase "if determined" or "if detecting (stated condition or event)" can be interpreted as "when determined" or "in response to determining" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)".

[0023] In addition, the sequence of steps in each of the following method embodiments is only an example and is not strictly limited.

[0024] In fact, the device deployed by the intelligent telescope scene image acquisition and processing system based on artificial intelligence of the present application can be flexibly configured according to the actual application scene. The system can be realized as: a hardware device running independently, an embedded module carried on an intelligent telescope, a cloud service instance or a virtual machine. For example, the system can be integrated in the intelligent telescope body as an embedded module to directly participate in image acquisition and real-time processing; or realized as a service instance deployed in a cloud server, to interact with multiple intelligent telescopes through wireless communication; or realized as a virtual machine, installed on an edge computing node, to balance the real-time processing and resource flexible scheduling requirements.

[0025] In terms of implementation, the system of the present application and the intelligent telescope terminal are mutually adapted. That is, the system is integrated in the intelligent telescope as an embedded module, to directly call the terminal hardware resources to complete acquisition and processing; the system is realized as a cloud service, to establish a connection with the intelligent telescope terminal through a wireless communication protocol such as 5G or WiFi; the system is realized as an edge computing node, to simultaneously interface with multiple adjacent intelligent telescopes, to provide low-latency data processing support.

[0026] In the embodiments of the present application, each module can be independently deployed and interacted through standardized interfaces. A certain module can interface with multiple associated modules and provide service support. Without modifying the core program code, the system application range can be expanded in the form of adding modules and expanding interfaces to realize cluster horizontal expansion and meet the needs of multi-device collaboration and large-scale data processing in different scenarios. In actual application, the above-mentioned modules can be centrally arranged in the same device, or distributed in different levels such as terminals, edge nodes, and clouds, and data transmission and instruction interaction can be completed through network communication.

[0027] Next, in combination with specific embodiments, each component of the system and the specific work flow are described in detail: Embodiment one, please refer to Figure 1 and Figure 2 As shown, the multi-lens cooperative acquisition and fusion module is used for assigning acquisition tasks for multiple intelligent telescopes, adapting the multi-terminal image feature data format and image communication protocol, and fusing to generate panoramic images or stereoscopic images.

[0028] In the embodiments of the present application, the multi-lens cooperative acquisition and fusion module includes an AI acquisition task scheduling unit, a multi-terminal data communication adaptation unit, and a multi-source image feature fusion unit, which cooperatively realize the functions of acquisition task assignment, data format unification, and image fusion.

[0029] The AI acquisition task scheduling unit is used for assigning image acquisition angles, spectral acquisition channels, and acquisition time sequences for each intelligent telescope in the distributed acquisition communication network.

[0030] Specifically, the AI acquisition task scheduling unit assigns acquisition angles based on the three-dimensional modeling data of the target scene, in combination with the current position coordinates and device performance parameters of multiple intelligent telescopes, and uses a greedy algorithm. The objective function of the algorithm is to maximize the acquisition field of view coverage efficiency and device load balancing degree, and to preferentially assign complex acquisition tasks to devices with stronger computing power. First, the coverage field of view angle of the panoramic image or stereoscopic image is determined, which is divided into a plurality of non-overlapping and edge-redundant sub-field of view angles, and then the corresponding sub-field of view angle acquisition tasks are assigned according to the lens parameters and resolution upper limit of each intelligent telescope.

[0031] Further, for spectral acquisition channel assignment, the required spectral band is determined according to the scene type, and in combination with the spectral acquisition capability of each intelligent telescope, the exclusive spectral channel or the time-sharing acquisition permission of the shared spectral channel is assigned to different devices. The acquisition time sequence assignment is based on the device response speed and communication delay, and uses a staggered acquisition strategy to avoid network congestion caused by multiple devices uploading data at the same time.

[0032] The multi-terminal data communication adaptation unit is used for unifying the image feature data format and image communication protocol output by multiple intelligent telescopes.

[0033] Specifically, the multi-terminal data communication adaptation unit pre-stores the image data format template and the communication protocol specification of the mainstream intelligent telescope. After receiving the image feature data uploaded by each device, the data header identification is first parsed to identify the original format and the communication protocol type.

[0034] Further, the image feature data of different formats is converted into a preset standardized format, and the pixel bit depth, channel order and feature descriptor structure are unified; the data transmitted by different communication protocols is protocol-converted and data-verified to ensure data integrity and consistency, thereby providing a unified data basis for subsequent multi-source image fusion.

[0035] The multi-source image feature fusion unit is configured to fuse the image feature data output by multiple intelligent telescopes into a panoramic image or a stereoscopic image through feature matching and pixel-level alignment.

[0036] In the embodiment of the present application, when the multi-source image feature fusion unit fuses the image feature data, it is specifically configured to: receive the image feature data output by multiple intelligent telescopes, the image feature data including the contour topology and texture association of the collection area of each device; perform feature matching on all the image feature data to establish a corresponding relationship between the image feature data of different devices; perform pixel-level alignment based on the corresponding relationship to eliminate the spatial offset of the image feature data; integrate the aligned image feature data to generate a panoramic image or a stereoscopic image.

[0037] Specifically, after receiving the image feature data output by multiple intelligent telescopes, each piece of data is first pre-processed to remove noise interference and invalid pixel points, and the key point coordinates of the contour topology and the gray value and gradient information of the texture association are retained.

[0038] Further, the ORB algorithm is used to perform feature matching on all the image feature data. This algorithm has both feature extraction efficiency and matching accuracy, and is suitable for the device computing power requirement on the adaptation side. Key feature points such as corners and edges in each image are extracted, the Euclidean distance between the feature points is calculated, the feature point pairs with a distance less than a preset threshold are determined as matching points, and a spatial corresponding relationship between the image data of different devices is established based on the matching points.

[0039] Further, pixel-level alignment is performed based on the corresponding relationship. The homography matrix transformation algorithm is used to take the image of one device as a reference to perform rotation, translation and scaling processing on the images of other devices, thereby eliminating spatial offset and distortion and ensuring accurate overlap of the matching points at the pixel level.

[0040] Further, the aligned image feature data is integrated, the pixel values in the overlapping area are processed by a weighted average fusion algorithm, the original pixel information in the non-overlapping area is directly retained, and a panoramic image or a stereoscopic image is generated according to application requirements.

[0041] In summary, the AI acquisition task scheduling unit can fully utilize the hardware performance of multiple intelligent telescopes by intelligently assigning acquisition angles, spectral channels and time sequences, avoid waste of overlapping acquisition areas or coverage blind spots, ensure the completeness and complementarity of the acquired data, and provide a basis for high-quality image fusion.

[0042] In summary, the multi-end data communication adaptation unit solves the data compatibility problem of different types of intelligent telescopes by unifying the data format and communication protocol, avoids fusion failure or image distortion caused by format differences, and ensures smooth flow of multi-source data and effectiveness of subsequent processing.

[0043] In summary, the multi-source image feature fusion unit can convert the acquisition data of multiple devices into panoramic or stereoscopic images through feature matching, pixel-level alignment and intelligent integration, break through the field of view and dimension limit of a single telescope, improve the field of view coverage range and spatial expressiveness of the image, and meet the application requirements of long-distance observation and large-scale scene recording.

[0044] Embodiment two, please refer to Figure 3 The intelligent telescope scene image acquisition and processing system based on artificial intelligence is shown in the figure, the distributed networking module is used to realize wireless interconnection of multiple intelligent telescopes, build a distributed acquisition communication network, and perform image communication fault redundancy processing.

[0045] In the embodiment of the application, the distributed networking module includes a distributed networking management unit and a communication fault redundancy processing unit, and each unit cooperates to realize network construction and fault tolerance functions.

[0046] The distributed networking management unit is used to realize wireless interconnection of multiple intelligent telescopes and build a distributed acquisition communication network.

[0047] Specifically, the distributed networking management unit uses self-organizing network technology, is based on 5G or WiFi6 wireless communication protocol, realizes automatic discovery and connection of multiple intelligent telescopes. Each intelligent telescope as a network node, through broadcasting its own device identifier, location information and communication capability, completes node discovery and identity authentication.

[0048] Further, based on the IEEE 802.11s protocol, a mesh network topology is adopted to construct a distributed acquisition communication network, and multi-path data transmission between nodes is realized. Each node can establish a communication link with multiple surrounding nodes to form a multi-path data transmission channel. After the network is constructed, a node connection state table is maintained in real time to record the communication quality, load condition and available link information of each node, thereby providing support for data transmission and fault handling.

[0049] A communication fault redundancy processing unit is configured to monitor the image communication state of each intelligent telescope in the distributed acquisition communication network, and when the image communication of any intelligent telescope is interrupted, a standby image communication link is switched and the surrounding intelligent telescopes are dispatched to complete image acquisition data.

[0050] In the embodiment of the present application, the communication fault redundancy processing unit is specifically used for: Continuously monitoring the image communication connection state of each intelligent telescope, and when it is detected that the image communication of a certain intelligent telescope is interrupted, starting a standby image communication link to attempt reconnection; If the reconnection fails, determining the preset image acquisition region and acquisition task of the intelligent telescope; Querying the idle intelligent telescopes within the communication coverage range of the faulty device in the distributed acquisition communication network, and assigning the idle intelligent telescopes with a complete image acquisition task; Receiving image feature data of the idle intelligent telescopes after completing the complete image acquisition task, which is used for subsequent image fusion.

[0051] Specifically, the communication connection state of each node is continuously monitored through a heartbeat packet mechanism, and the heartbeat packet sending period is set to 100 ms. If no response packet is received from a certain node for 3 consecutive periods, it is determined that the image communication of the node is interrupted. At this time, a standby communication link is automatically started to attempt to reestablish the connection.

[0052] Further, if the standby link reconnection fails, the preset acquisition region coordinates, field of view range and acquisition task parameters of the faulty device are retrieved from the network node state table. At the same time, the node state within the communication coverage range of the faulty device is queried, and within a radius of 500 meters centered on the position of the faulty device, idle intelligent telescopes that currently have no acquisition task or have a load lower than 30% are selected.

[0053] Further, according to the position and performance of the idle device, the nearest principle is adopted to assign a complete acquisition task, and the consistency requirement of the complete acquisition region and the original acquisition task is specified. After receiving the task, the idle device performs acquisition according to the preset parameters, and after completion, uploads the image feature data to the distributed networking management unit for subsequent fusion processing.

[0054] In summary, the distributed networking management unit constructs a distributed acquisition and communication network through self-organizing network technology, enabling flexible interconnection and multi-path data transmission among multiple smart telescopes, improving network scalability and stability, and providing reliable communication support for multi-telescope collaborative acquisition.

[0055] In summary, the communication fault redundancy processing unit, through a dual redundancy mechanism of link switching and task re-acquisition, can quickly respond to equipment communication interruption problems, avoid data loss, ensure the continuity and integrity of image acquisition, and improve the reliability of the system in complex communication environments.

[0056] Example 3, please refer to Figure 4 As shown, the AI-based intelligent telescope scene image acquisition and processing system uses a content-aware compression and transmission module to identify the importance of scene image regions and, in conjunction with the wireless communication link status, to perform differentiated compression of image data and adjustment of transmission parameters.

[0057] In this embodiment of the invention, the content-aware compression transmission module includes an image content importance identification unit, a dynamic compression strategy generation unit, a communication link status monitoring unit, and a transmission parameter adaptive adjustment unit. These units work together to achieve content identification, compression strategy generation, and transmission optimization functions. The image content importance recognition unit is used to identify the importance of different regions in a scene image.

[0058] Specifically, a deep learning semantic segmentation model is used to divide the scene image into regions, identifying the core scene region and the background region in the image. By calculating the information entropy and edge density of each region, the importance of the region is quantified: the core scene region with an information entropy higher than 8 bits and an edge density higher than 0.6 is identified as a high-importance region; the background region with an information entropy lower than 4 bits and an edge density lower than 0.3 is identified as a low-importance region; and the remaining regions are identified as medium-importance regions.

[0059] The dynamic compression strategy generation unit is used to generate differentiated compression strategies based on regional importance.

[0060] Specifically, for highly important areas, the lossless compression algorithm LZ77 or a lossy compression algorithm with a compression ratio of ≤2:1 is used to ensure that the detailed information of the core scene is not lost; for medium-important areas, a lossy compression algorithm with a compression ratio of 2:1 < ≤5:1 is used to reduce the amount of data while retaining key information; for low-important areas, a lossy compression algorithm with a compression ratio of >5:1 is used to minimize the amount of data transmitted. The communication link status monitoring unit is used to monitor the signal strength and stability of the wireless communication link.

[0061] Specifically, the signal strength indicator, signal-to-noise ratio (SNR), and packet loss rate (PLR) of the communication link are collected in real time. The signal strength threshold is set to -70dBm, the SNR threshold to 20dB, and the packet loss rate threshold to 5%. When RSSI ≥ -70dBm, SNR ≥ 20dB, and PLR ≤ 5%, the link is considered to be in good condition; when any of these indicators exceeds the threshold, the link is considered to be in poor condition.

[0062] The transmission parameter adaptive adjustment unit is used to adjust the transmission frame rate of image data packets based on the communication link status and compression processing results.

[0063] Specifically, when the link is in good condition, the transmission frame rate is adjusted to 15-30fps based on the amount of compressed data to ensure real-time transmission; when the link deteriorates, the transmission frame rate is reduced to 5-15fps to reduce data transmission pressure, while prioritizing the transmission of image data packets in important areas to ensure the effective transmission of core information.

[0064] In summary, the image content importance recognition unit can accurately distinguish between the core and background regions of an image through semantic segmentation and quantitative evaluation, providing a basis for differentiated processing and avoiding the loss of core information caused by indiscriminate compression.

[0065] In summary, the dynamic compression strategy generation unit formulates differentiated compression schemes based on regional importance, maximizing the reduction of data volume, improving transmission efficiency, and saving communication bandwidth resources while ensuring the quality of core images.

[0066] In summary, the communication link status monitoring unit can promptly grasp the link operation status through real-time monitoring of multiple indicators, providing accurate reference for adjusting transmission parameters and avoiding transmission interruptions or data loss caused by changes in link status.

[0067] In summary, the transmission parameter adaptive adjustment unit dynamically adjusts the transmission frame rate based on the link status and compression results, achieving a balance between transmission efficiency and image quality, and ensuring stable transmission of critical image information under different communication environments.

[0068] Example 4, please refer to Figure 5 As shown, the AI-based intelligent telescope scene image acquisition and processing system has a feature encryption and isolation module used to extract abstract features from scene images, identify the sensitivity of the images, and process the abstract features using an encryption method that matches the sensitivity.

[0069] In this embodiment of the invention, the feature encryption isolation module includes a lightweight feature extraction unit on the edge and a hierarchical encryption unit on the edge, and each unit works together to achieve feature extraction and hierarchical encryption functions.

[0070] The edge-side lightweight feature extraction unit is used to extract abstract features of scene images, including contour topology and texture association.

[0071] In this embodiment of the invention, when the end-side lightweight feature extraction unit extracts abstract features of a scene image, it is specifically used for: Pixel-level feature analysis is performed on scene images to separate contour topology-related features from texture-related features; The contour topology-related features are extracted in a structured manner to form a contour topology. The texture association-related features are analyzed to form a texture association. The contour topology and texture association are integrated to obtain the abstract features of the scene image.

[0072] Specifically, a lightweight convolutional neural network, MobileNetV3, is used to perform pixel-level feature analysis on scene images, balancing feature extraction accuracy with edge computing power consumption. Low-level features such as edges and corners are extracted using 3×3 convolutional kernels, and key information is retained through pooling operations, separating shape features related to contour topology and surface features related to texture.

[0073] Furthermore, structured extraction of contour topology features is performed, using Hough transform to detect contour elements such as lines and curves, establishing spatial coordinate relationships between key contour points, and forming a structured contour topology. Correlation analysis is conducted on texture-related features, calculating the gray-level difference and correlation between adjacent pixels to construct a texture feature matrix and form texture associations. Finally, the coordinate matrix of the contour topology and the feature matrix of the texture associations are dimensionally aligned and integrated to obtain the abstract features of the scene image.

[0074] The edge-side hierarchical encryption unit is used to identify the scene type of the image, determine the image sensitivity, and process the extracted abstract features using a hierarchical encryption method that matches the sensitivity.

[0075] In this embodiment of the invention, when the end-side hierarchical encryption unit identifies the sensitivity of an image and processes abstract features, it is specifically used for: Perform scene semantic recognition on scene images to determine whether they contain human information or confidential markings, and determine the sensitivity of the images; Asymmetric encryption algorithms are used for highly sensitive images containing human information or classified identifiers, while symmetric encryption algorithms are used for low-sensitivity images that do not contain such information. The extracted abstract features are encrypted to generate encrypted abstract features.

[0076] Specifically, a pre-trained scene recognition model is used to perform scene semantic recognition on scene images, detecting whether the image contains human body information such as human contours and facial features, or contains confidential markings. If the above information is detected, it is determined to be a highly sensitive image; if not detected, it is determined to be a low-sensitivity image.

[0077] Furthermore, for the abstract features of highly sensitive images, an asymmetric encryption algorithm is used for encryption, employing public-key encryption and private-key decryption to ensure data security. For the abstract features of low-sensitivity images, a symmetric encryption algorithm is used, improving encryption and decryption efficiency while ensuring basic security. During the encryption process, the abstract feature data is fragmented into 1KB to 2KB chunks, each chunk is independently encrypted and then a CRC32 checksum is added to ensure data integrity during transmission and storage, generating the encrypted abstract feature data.

[0078] In summary, the edge-side lightweight feature extraction unit uses a lightweight network and structured extraction method to efficiently extract abstract features on the edge, avoiding bandwidth consumption and latency caused by the transmission of a large amount of original image data, while retaining the core feature information of the image, providing a foundation for subsequent edge-cloud collaborative processing.

[0079] In summary, the edge-side hierarchical encryption unit adopts differentiated encryption methods based on the sensitivity of images, which not only ensures the privacy and security of highly sensitive image data, but also takes into account the processing efficiency of low-sensitivity images, avoids resource waste caused by over-encryption, and improves the overall operating efficiency of the system.

[0080] Example 5, please refer to Figure 6 As shown, the intelligent telescope scene image acquisition and processing system based on artificial intelligence has an end-to-cloud collaborative communication module used to complete the image reconstruction, model iteration and image communication protocol switching required for end-to-cloud image communication based on encrypted abstract features.

[0081] In this embodiment of the invention, the edge-cloud collaborative communication module includes a cloud feature reconstruction optimization unit, a federated learning model iteration unit, and a communication protocol dynamic adaptation unit. Each unit works together to realize image reconstruction, model update, and protocol adaptation functions.

[0082] The cloud-based feature reconstruction and optimization unit is used to receive encrypted abstract features and complete image reconstruction based on the abstract features.

[0083] Specifically, after the encrypted abstract features are uploaded to the cloud receiving end, they are decrypted using the corresponding decryption algorithm to obtain the original abstract features. RSA keys are used for highly sensitive images, and AES keys are used for low-sensitivity images. Based on the contour topological coordinate matrix in the abstract features, a contour reconstruction algorithm is used to generate the basic framework of the image. Then, combined with the texture association feature matrix, a texture mapping algorithm is used to fill in the image details, resulting in a preliminary reconstructed image.

[0084] Furthermore, an image super-resolution algorithm is used to optimize the initially reconstructed image. The algorithm iteration count is set to 200-300 times, and the generator learning rate is 0.0001, which improves the image resolution and detail representation, and finally generates a clear and complete reconstructed image.

[0085] The federated learning model iteration unit is used to receive model data uploaded from the edge, aggregate the model data to update the global image processing algorithm, and feed the updated global image processing algorithm back to the edge.

[0086] In this embodiment of the invention, when the federated learning model iterative unit aggregates model data and updates the algorithm, it is specifically used for: It receives model data uploaded from multiple smart telescopes, and the model data is associated with the abstract features of the scene images collected by each telescope. The confidence weights for each end-side model data are calculated using the following formula: ; In the formula, W i Q represents the credibility weight of the data from the i-th intelligent telescope end-side model. i Let F be the data acquisition reliability coefficient at the i-th intelligent telescope end. i Let Q be the abstract feature quality coefficient corresponding to the gradient data uploaded by the i-th intelligent telescope end, and n be the total number of intelligent telescope ends participating in federated learning in the distributed acquisition and communication network. j Let F be the data acquisition reliability coefficient at the j-th intelligent telescope end. j The abstract feature quality coefficient corresponding to the gradient data uploaded to the j-th intelligent telescope end-side; All model data are weighted and aggregated based on credibility weights to generate aggregated gradient data. The parameters of the global image processing algorithm are updated based on the aggregated gradient data to obtain the updated global image processing algorithm. The updated global image processing algorithm is fed back to each smart telescope end.

[0087] Specifically, it receives model data uploaded from each terminal, including gradient data and loss function values ​​of the terminal image processing model, and the model data is associated with the abstract features collected from each terminal to ensure data traceability.

[0088] Furthermore, calculate the reliability weight of the model data at each end: the data acquisition reliability coefficient Q. i The coefficient is determined based on indicators such as historical acquisition success rate and data integrity of the edge device, with a value range of 0.5 to 1.0. The better the acquisition performance, the closer the coefficient is to 1.0; the abstract feature quality coefficient F i The evaluation is based on indicators such as feature clarity and completeness, with values ​​ranging from 0.5 to 1.0. The higher the feature quality, the closer the coefficient is to 1.0. The credibility weight of each edge-side model data is calculated using the above formula, and the sum of the weights is 1.0.

[0089] Furthermore, all model data are weighted and aggregated based on confidence weights. The gradient data from each endpoint is multiplied by its corresponding weight and then summed to generate aggregated gradient data. The stochastic gradient descent algorithm is used to update the parameters of the global image processing algorithm based on the aggregated gradient data. The algorithm is iteratively optimized until the loss function value is lower than a preset threshold. This threshold can be set to 0.001 to 0.01 depending on the actual application scenario, resulting in the updated global image processing algorithm.

[0090] Furthermore, the updated global image processing algorithm is fed back to each smart telescope end via an encrypted communication channel. After receiving the algorithm, the end replaces the original algorithm, thereby realizing dynamic iterative optimization of the model.

[0091] The communication protocol dynamic adaptation unit is used to monitor the communication link status and switch the image communication protocol based on the communication link status.

[0092] Specifically, the system monitors communication link metrics such as bandwidth, latency, and packet loss rate in real time. When the link bandwidth is ≥10Mbps, latency is ≤50ms, and packet loss rate is ≤2%, the HTTP / 2 protocol is used for image data transmission to improve transmission efficiency. When the link bandwidth is <10Mbps, latency is >50ms, or packet loss rate is >2%, the system automatically switches to the MQTT protocol, using a lightweight data transmission mode to reduce link occupancy and transmission latency. When the link bandwidth is <1Mbps or packet loss rate is >10%, the system switches to the CoAP protocol, transmitting only key abstract feature data to ensure effective interaction of core information.

[0093] In summary, the cloud-based feature reconstruction and optimization unit completes image reconstruction and optimization based on abstract features. It can restore high-quality images while only transmitting feature data on the device side, thus saving bandwidth and ensuring image output quality.

[0094] In summary, the federated learning model iteration unit achieves secure and efficient fusion of multi-device model data through credibility weight calculation and gradient aggregation, avoiding the privacy leakage risks caused by centralized data collection. At the same time, it enables the global algorithm to adapt to the actual collection scenarios of each device, improving the algorithm's versatility and processing accuracy.

[0095] In summary, the communication protocol dynamic adaptation unit flexibly switches communication protocols based on link status, enabling efficient data transmission in different network environments. This avoids poor transmission adaptability caused by fixed protocols and improves the stability and flexibility of end-to-cloud collaborative communication.

[0096] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0097] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0098] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. An intelligent telescope scene image acquisition and processing system based on artificial intelligence, characterized in that, The system includes a multi-camera collaborative acquisition and fusion module, a distributed networking module, a content-aware compression and transmission module, a feature-based encryption and isolation module, and an end-to-cloud collaborative communication module. The multi-mirror collaborative acquisition and fusion module assigns acquisition tasks to multiple smart telescopes, adapts the image feature data formats and image communication protocols of multiple terminals, and fuses them to generate panoramic or stereoscopic images. The distributed networking module enables wireless interconnection of multiple smart telescopes, constructs a distributed acquisition and communication network, and performs image communication fault redundancy processing. The content-aware compression transmission module identifies the importance of scene image regions and, in conjunction with the wireless communication link status, completes differentiated compression of image data and adjusts transmission parameters. The feature encryption and isolation module extracts abstract features from the scene image, identifies the image sensitivity, and processes the abstract features using an encryption method that matches the sensitivity. The edge-cloud collaborative communication module completes the image reconstruction, model iteration, and image communication protocol switching required for edge-cloud image communication based on encrypted abstract features.

2. The intelligent telescope scene image acquisition and processing system based on artificial intelligence according to claim 1, characterized in that, The multi-mirror collaborative acquisition and fusion module includes an AI acquisition task scheduling unit, a multi-terminal data communication adaptation unit, and a multi-source image feature fusion unit. The AI ​​acquisition task scheduling unit allocates image acquisition angles, spectral acquisition channels, and acquisition timing to each intelligent telescope in the distributed acquisition and communication network. The multi-terminal data communication adapter unit unifies the image feature data format and image communication protocol output by multiple smart telescopes. The multi-source image feature fusion unit fuses image feature data from multiple smart telescopes into panoramic or stereoscopic images through feature matching and pixel-level alignment. When the multi-source image feature fusion unit fuses image feature data, it is specifically used for: Receive image feature data output from multiple smart telescopes, wherein the image feature data includes the contour topology and texture association of the acquisition area of ​​each device; Perform feature matching on all image feature data to establish the correspondence between image feature data from different devices; Pixel-level alignment is performed based on the correspondence to eliminate spatial offset of image feature data; Integrate the aligned image feature data to generate panoramic or stereoscopic images.

3. The intelligent telescope scene image acquisition and processing system based on artificial intelligence according to claim 1, characterized in that, The distributed networking module includes a distributed networking management unit and a communication fault redundancy processing unit. The distributed networking management unit enables wireless interconnection of multiple smart telescopes to construct a distributed data acquisition and communication network. The communication fault redundancy processing unit monitors the image communication status of each smart telescope in the distributed acquisition and communication network. When the image communication of any smart telescope is interrupted, it switches to the backup image communication link and schedules surrounding smart telescopes to complete the image acquisition data.

4. The intelligent telescope scene image acquisition and processing system based on artificial intelligence according to claim 3, characterized in that, When the communication fault redundancy processing unit handles image communication faults, it is specifically used for: The image communication connection status of each smart telescope is continuously monitored. When an image communication interruption is detected in a smart telescope, a backup image communication link is activated to attempt to reconnect. If reconnection fails, determine the preset image acquisition area and acquisition task of the smart telescope; In the distributed acquisition and communication network, query the idle smart telescopes within the communication coverage area of ​​the faulty device, and assign image completion acquisition tasks to the idle smart telescopes. The system receives image feature data after the idle smart telescope completes the image acquisition task, and uses it for subsequent image fusion.

5. The intelligent telescope scene image acquisition and processing system based on artificial intelligence according to claim 1, characterized in that, The content-aware compression transmission module includes an image content importance recognition unit, a dynamic compression strategy generation unit, a communication link status monitoring unit, and a transmission parameter adaptive adjustment unit. The image content importance recognition unit identifies the importance of different regions in the scene image; The dynamic compression strategy generation unit generates differentiated compression strategies based on regional importance; The communication link status monitoring unit monitors the signal strength and stability of the wireless communication link; The adaptive transmission parameter adjustment unit adjusts the transmission frame rate of image data packets based on the communication link status and compression processing results.

6. The intelligent telescope scene image acquisition and processing system based on artificial intelligence according to claim 1, characterized in that, The feature encryption isolation module includes a lightweight feature extraction unit on the edge and a hierarchical encryption unit on the edge. The edge-side lightweight feature extraction unit extracts abstract features of the scene image, including contour topology and texture association; The edge-side hierarchical encryption unit identifies the scene type of the image, determines the image sensitivity, and processes the extracted abstract features using a hierarchical encryption method that matches the sensitivity.

7. The intelligent telescope scene image acquisition and processing system based on artificial intelligence according to claim 6, characterized in that, When the edge-side lightweight feature extraction unit extracts abstract features of the scene image, it is specifically used for: Pixel-level feature analysis is performed on scene images to separate contour topology-related features from texture-related features; The contour topology-related features are extracted in a structured manner to form a contour topology. The texture association-related features are analyzed to form a texture association. The contour topology and texture association are integrated to obtain the abstract features of the scene image.

8. The intelligent telescope scene image acquisition and processing system based on artificial intelligence according to claim 6, characterized in that, When the end-side hierarchical encryption unit identifies image sensitivity and processes abstract features, it is specifically used for: Perform scene semantic recognition on scene images to determine whether they contain human information or confidential markings, and determine the sensitivity of the images; Asymmetric encryption algorithms are used for highly sensitive images containing human information or classified identifiers, while symmetric encryption algorithms are used for low-sensitivity images that do not contain such information. The extracted abstract features are encrypted to generate encrypted abstract features.

9. The intelligent telescope scene image acquisition and processing system based on artificial intelligence according to claim 1, characterized in that, The edge-cloud collaborative communication module includes a cloud feature reconstruction and optimization unit, a federated learning model iteration unit, and a communication protocol dynamic adaptation unit. The cloud-based feature reconstruction and optimization unit receives encrypted abstract features and completes image reconstruction based on the abstract features; The federated learning model iteration unit receives model data uploaded from the end side, aggregates the model data to update the global image processing algorithm, and feeds back the updated global image processing algorithm to the end side. The communication protocol dynamic adaptation unit monitors the communication link status and switches the image communication protocol based on the communication link status.

10. The intelligent telescope scene image acquisition and processing system based on artificial intelligence according to claim 9, characterized in that, When the federated learning model iterative unit aggregates model data and updates the algorithm, it is specifically used for: The system receives model data uploaded from multiple smart telescopes, and the model data is associated with the abstract features of the scene images collected by each telescope. The confidence weights for each end-side model data are calculated using the following formula: ; In the formula, W i Q represents the credibility weight of the data from the i-th intelligent telescope end-side model. i Let F be the data acquisition reliability coefficient at the i-th intelligent telescope end. i Let Q be the abstract feature quality coefficient corresponding to the gradient data uploaded by the i-th intelligent telescope end, and n be the total number of intelligent telescope ends participating in federated learning in the distributed acquisition and communication network. j Let F be the data acquisition reliability coefficient at the j-th intelligent telescope end. j The abstract feature quality coefficient corresponding to the gradient data uploaded to the j-th intelligent telescope end-side; Based on the aforementioned confidence weights, all model data are weighted and aggregated to generate aggregated gradient data. The parameters of the global image processing algorithm are updated based on the aggregated gradient data to obtain the updated global image processing algorithm. The updated global image processing algorithm is fed back to each smart telescope end.

Images