Low-light night vision optical performance test system based on data modeling

Abstract

The invention discloses a data modeling-based low-light night vision optical performance test system, and relates to the technical field of photoelectric instrument performance detection and intelligent diagnosis, and the system comprises a target plate module which is used for providing an observation target with a division pattern for a detected low-light night vision instrument; and the image acquisition and processing module is used for acquiring an image output after the detected low-light night vision instrument observes the target plate module, and processing the image to extract image feature data. According to the data modeling-based low-light night vision optical performance test system, through automatic image acquisition and processing and in combination with a multi-algorithm parallel computing and cross verification process, a traditional method depending on visual judgment of an operator is replaced, the influence of subjective factors and environmental fluctuation on a detection result is effectively reduced, and the detection accuracy is improved. And the consistency and objectivity of the detection result are improved. The system can automatically complete the whole process from image acquisition, feature extraction to parameter verification, and the efficiency and reliability of detection operation are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of performance detection and intelligent diagnosis of photoelectric instruments, in particular to a low-light-level night vision optical performance test system based on data modeling. BACKGROUND

[0002] Traditional optical performance detection methods mainly rely on a parallel light tube to simulate an infinite target, combined with high-precision optical components and a special laboratory environment, to realize detection of core indicators such as instrument resolution and modulation transfer function (MTF). Although such methods have high detection accuracy, they have the disadvantages of large equipment size, poor environmental adaptability, high cost, and difficulty in rapid deployment and real-time detection in the field or with accompanying support conditions. To meet the needs of field detection, some portable or limited-distance detection schemes have been proposed in existing technologies. For example, the authorized publication "CN114993629B" describes a "visible light low-light-level night vision instrument field optical performance detection method". This method builds a detection system including a target plate adjustment module, a visible light / low-light-level target plate, a laser ranging module, and a light reduction cover, and completes the resolution detection of photoelectric instruments in limited distance combined with target plates. This system discards the traditional parallel light tube, has the advantages of small size, light weight, no need for external power supply, easy to carry and quickly deploy, etc., especially through the design of the light reduction cover, it realizes the detection of low-light-level instruments in strong light conditions during the day, significantly expanding the detection opportunity and applicable environment.

[0003] However, this detection method still has the following shortcomings: first, the detection process relies on the operator to visually judge the clarity of the target plate pattern subjectively, and to determine the resolution value by referring to the lookup table, which is affected by human experience, visual fatigue, and environmental light fluctuations, and the consistency, repeatability, and objectivity need to be improved; second, the system lacks the ability to automatically collect, process, and analyze detection data, and cannot realize digital recording, performance trend analysis, and intelligent diagnosis of the detection process; third, although this method is suitable for field rapid detection, it does not build a performance evaluation model associated with the instrument's use environment and working conditions, making it difficult to realize instrument health state prediction and maintenance decision support based on historical data and multi-parameter fusion.

[0004] With the development of intelligent and high-precision photoelectric instruments, and the in-depth application of big data and artificial intelligence technology in equipment support, it is an urgent need to build a low-light-level night vision optical performance test system that can integrate data collection, modeling analysis, and intelligent evaluation, to improve detection efficiency and realize state prediction and precise support. SUMMARY

[0005] The present application aims to provide a low-light-level night vision optical performance test system based on data modeling to solve the problems raised in the background.

[0006] To solve the above technical problems, the present application provides the following technical solutions: a micro-light night vision optical performance test system based on data modeling, comprising:

[0007] A target plate module is used to provide an observation target with a reticle pattern for a micro-light night vision instrument under test;

[0008] An image acquisition and processing module is used to acquire an image output by the micro-light night vision instrument under test after observing the target plate module, and process the image to extract image feature data;

[0009] An environmental parameter acquisition module is used to acquire environmental data around the target plate module in real time;

[0010] A data processing and analysis host is connected to the image acquisition and processing module and the environmental parameter acquisition module, respectively;

[0011] The data processing and analysis host comprises:

[0012] An image sharpness evaluation and verification unit is used to receive the image feature data, and perform parallel calculation and cross verification on the image feature data based on a plurality of preset image sharpness evaluation algorithms, and output verified optical performance parameters;

[0013] A performance modeling and diagnosis unit is used to receive the verified optical performance parameters and the environmental data, analyze based on a pre-trained time series prediction model, and generate performance state diagnosis results and trend prediction information of the micro-light night vision instrument under test;

[0014] A data closed-loop feedback unit is used to receive the diagnosis results output by the performance modeling and diagnosis unit, and dynamically adjust the processing parameters of the image acquisition and processing module or the image sharpness evaluation and verification unit according to the diagnosis results.

[0015] Further, the image sharpness evaluation and verification unit performs the following steps for verification:

[0016] S11: receiving the image feature data from the image acquisition and processing module, wherein the image feature data at least includes a first feature value calculated based on an edge gradient algorithm, and a second feature value calculated based on an image signal-to-noise ratio algorithm;

[0017] S12: comparing the first feature value with a first preset threshold range, and comparing the second feature value with a second preset threshold range, if both are within the respective preset threshold range, determining that the image quality is qualified, and executing step S13; if any one is not within the corresponding preset threshold range, determining it as abnormal data, and sending an instruction to the image acquisition and processing module to trigger secondary acquisition and processing of the current target plate image.

[0018] S13: calling no less than three different image sharpness evaluation algorithms to perform parallel calculation on the image feature data qualified in quality, and obtaining no less than three intermediate evaluation results;

[0019] S14: weighting and fusing the no less than three intermediate evaluation results, and matching the fusion result with a preset resolution lookup table to obtain a preliminary optical performance parameter value;

[0020] S15: comparing the preliminary optical performance parameter value with a historical parameter interval of the same type of instrument in a historical detection database, if the preliminary parameter value deviates from the historical parameter interval, marking as data to be reviewed, and repeating steps S12 to S14 for review;

[0021] S16: outputting the data reviewed or confirmed as normal by step S15 to the performance modeling and diagnosis unit as the verified optical performance parameter.

[0022] Further, the pre-trained time series prediction model in the performance modeling and diagnosis unit is a long short-term memory neural network model, and its construction and operation include:

[0023] The input feature vector received by the input layer of the model includes: the verified optical performance parameter, the temperature data and humidity data collected by the environment parameter acquisition module in real time, and the cumulative working time data of the measured low-light night vision instrument;

[0024] The hidden layer of the model contains a plurality of memory units for learning and memorizing the change pattern and dependency relationship of the input feature vector in the time dimension;

[0025] The output of the output layer of the model includes: the optical performance parameter prediction value of the measured low-light night vision instrument in the next detection period, the health index indicating the performance degradation degree, and the time node information of the recommended maintenance operation.

[0026] Further, the performance modeling and diagnosis unit further includes a model updating module for performing iterative optimization of the model:

[0027] The model updating module stores the verified optical performance parameter, the corresponding environment data and the diagnosis result obtained each time in the local historical database;

[0028] The model updating module periodically re-trains the long short-term memory neural network model using new data in the local historical database to update the internal weight parameters of the model;

[0029] The system supports data synchronization with a cloud server, which aggregates historical data of multiple test systems of the same model to generate and distribute globally optimized model parameters to the local system.

[0030] Further, the process of adjusting parameters according to the performance state diagnosis result includes:

[0031] When the diagnosis result indicates that the performance of the tested low-light night vision instrument is in a stable state, the data closed-loop feedback unit keeps the current processing parameters of the image acquisition and processing module unchanged;

[0032] When the diagnosis result indicates that the performance of the tested low-light night vision instrument fluctuates or declines, the data closed-loop feedback unit sends instructions to the image clarity evaluation and verification unit to adjust the algorithm weight used when performing weighted fusion, and increase the weight proportion of the algorithm more sensitive to image noise;

[0033] The data closed-loop feedback unit also dynamically adjusts the working gain of the image sensor in the image acquisition and processing module according to the light intensity information in the environmental data.

[0034] Further, the image acquisition and processing module includes an image sensor and a signal processing circuit, the image sensor is optically coupled to the rear end of the objective lens of the tested low-light night vision instrument, for directly capturing the optical image output by the objective lens and converting it into a digital electrical signal; the signal processing circuit is used for preprocessing the digital electrical signal, including analog-to-digital conversion, noise filtering and illumination compensation, to generate the image feature data.

[0035] Further, the target plate module includes a support frame and a target plate detachably mounted on the top of the support frame, the front and back surfaces of the target plate are coated with different reflectivity diffuse reflection coatings, respectively forming high-contrast observation surfaces and low-contrast observation surfaces; the environmental parameter acquisition module is integrated on the support frame, including temperature sensors, humidity sensors and illuminance sensors.

[0036] Further, the data processing and analysis host also includes a human-computer interaction and display unit, which is used for:

[0037] Real-time display of the original image collected by the image acquisition and processing module, the processed image feature data;

[0038] Visualize the history curve of the verified optical performance parameters, the performance trend prediction curve and the health index generated by the performance modeling and diagnosis unit in the form of charts;

[0039] Receive user input operation instructions and set system alarm thresholds.

[0040] Further, the workflow of the system comprises:

[0041] Step a: set the target plate module and the measured low-light night vision instrument at a preset detection distance, start the measured low-light night vision instrument and make it aim at the target plate module;

[0042] Step b: the image acquisition and processing module acquires images and performs preprocessing, and the environment parameter acquisition module acquires environment data, and both send the data to the data processing and analysis host;

[0043] Step c: the image definition evaluation and verification unit performs multiple rounds of parallel calculation and cross verification on the received data, and outputs the verified optical performance parameters after rechecking and confirmation;

[0044] Step d: the performance modeling and diagnosis unit combines the environment data, analyzes the verified optical performance parameters by using the time series prediction model, and generates the diagnosis result containing prediction information;

[0045] Step e: the data closed-loop feedback unit adjusts the related processing parameters in the subsequent detection of the system according to the diagnosis result.

[0046] Further, in the step c, the number of times of the multiple rounds of parallel calculation and cross verification is not less than two, and each verification is based on an independent algorithm combination or parameter setting; after the step e, the system archives the complete data packet of this detection, including the original data, intermediate processing data, final result and adjusted parameters, to the local historical database for subsequent model updating and analysis tracing.

[0047] The application provides a low-light night vision optical performance test system based on data modeling.

[0048] The low-light night vision optical performance test system based on data modeling replaces the traditional method of relying on the visual judgment of operators by automatic image acquisition and processing, combined with the multiple algorithm parallel calculation and cross verification process, effectively reduces the influence of subjective factors and environmental fluctuations on the detection result, and improves the consistency and objectivity of the detection result. The system can automatically complete the whole process from image acquisition, feature extraction to parameter verification, improve the efficiency and reliability of the detection operation.

[0049] The micro-light night vision optical performance test system based on data modeling fuses and analyzes the verified performance parameters and environmental data by using a time series prediction model, realizes quantitative evaluation and prediction of the instrument performance degradation trend, and forms a closed-loop feedback based on the diagnosis result to dynamically adjust the front-end processing parameters. This not only extends the performance evaluation from single detection to long-term state monitoring, but also provides data support for preventive maintenance, and enhances the adaptive ability and intelligent level of the test system. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 The system architecture diagram of the micro-light night vision optical performance test system based on data modeling of the application is shown in the figure.

[0051] Figure 2 The image sharpness evaluation verification flowchart of the micro-light night vision optical performance test system based on data modeling of the application is shown in the figure. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the application.

[0053] Please refer to Figure 1 and Figure 2 The application provides a technical solution: a micro-light night vision optical performance test system based on data modeling, comprising:

[0054] A target plate module is configured to provide an observation target with a reticle pattern for a micro-light night vision instrument under test.

[0055] An image acquisition and processing module is configured to acquire images output by the micro-light night vision instrument under test after observing the target plate module, and process the images to extract image feature data.

[0056] An environmental parameter acquisition module is configured to acquire environmental data around the target plate module in real time.

[0057] A data processing and analysis host is connected to the image acquisition and processing module and the environmental parameter acquisition module.

[0058] The data processing and analysis host comprises:

[0059] An image sharpness evaluation and verification unit is configured to receive the image feature data, and perform parallel calculation and cross verification on the image feature data based on a plurality of preset image sharpness evaluation algorithms, and output verified optical performance parameters.

[0060] a performance modeling and diagnosis unit for receiving the verified optical performance parameters and environmental data, performing analysis based on a pre-trained time series prediction model, and generating performance state diagnosis results and trend prediction information of the tested low-light-level night vision instrument;

[0061] a data closed-loop feedback unit for receiving the diagnosis results output by the performance modeling and diagnosis unit, and dynamically adjusting the processing parameters of the image acquisition and processing module or the image sharpness evaluation and verification unit according to the diagnosis results.

[0062] It should be further explained that the system comprises multiple functional modules, wherein the target plate module is composed of a support frame and a replaceable target plate, the surface of the target plate is coated with a diffuse reflection coating with a specific reflectivity and etched with a standard scale pattern, and is used to provide an observation target for the low-light-level night vision instrument to be tested. The image acquisition and processing module is fixedly connected to the rear of the eyepiece of the low-light-level night vision instrument through an adapter interface, and the core of the image acquisition and processing module is a high-sensitivity image sensor that directly receives the optical image from the eyepiece and converts it into a digital signal. The digital signal is then preprocessed by a processing circuit comprising an analog-to-digital converter, a programmable gain amplifier, and a digital filter, to generate image feature data including edge, contrast, and noise level.

[0063] The environmental parameter acquisition module is integrated in the support frame of the target plate module and comprises temperature, humidity, and illumination sensors for synchronously acquiring environmental data of the test site. The data processing and analysis host is connected to the above-mentioned modules through wired or wireless means, and the image sharpness evaluation and verification unit running inside the data processing and analysis host adopts a multi-mechanism parallel processing and cross-validation workflow: first, edge sharpness calculations based on Sobel, Canny, and Laplace operators are performed on the incoming image feature data, and the signal-to-noise ratio of the image is also calculated; then, a decision logic is used to compare the results of each algorithm with the preset qualified threshold interval, and if the results are abnormal, the image acquisition module is automatically triggered to perform reacquisition; for data that passes the initial screening, the unit calls at least three different sharpness evaluation algorithms for independent operation, and the multiple intermediate results obtained are fused according to preset weight coefficients, and the fused results are used to query a pre-stored resolution lookup table to obtain the optical performance parameters; finally, the parameters are compared with the historical data interval of the same type of instrument, and if they deviate from the typical range, a secondary review process is started to ensure the reliability of the output results.

[0064] The performance modeling and diagnosis unit is built-in a pre-trained long short-term memory neural network model, the input of which is the verified optical performance parameter sequence, the environmental parameter sequence and the cumulative working time of the instrument, the model learns the correlation and evolution rule of these parameters in the time dimension through its internal memory gate structure, and then outputs the prediction value of future performance change, the health score representing the comprehensive state of the instrument and the maintenance opportunity suggestion.

[0065] The data closed-loop feedback unit dynamically adjusts the system parameters according to the diagnosis conclusion output by the performance modeling unit, for example, when the diagnosis prompts that the instrument performance appears a slight degradation trend, the unit will automatically adjust the weight distribution of different algorithms in the image sharpness evaluation and verification unit, or send instructions to the image acquisition and processing module to fine-tune its gain and filter parameters, so that the whole system has adaptive optimization capability.

[0066] The modules work together to realize a complete closed-loop test process from automatic image acquisition, multi-round algorithm verification, performance evaluation and prediction based on time series model, to feedback and optimization of acquisition and processing parameters according to the evaluation results, effectively improving the objectivity, consistency and intelligent level of detection.

[0067] The image sharpness evaluation and verification unit performs the following steps for verification:

[0068] S11: receiving image feature data from the image acquisition and processing module, the image feature data at least including a first feature value calculated based on an edge gradient algorithm and a second feature value calculated based on an image signal-to-noise ratio algorithm;

[0069] S12: comparing the first feature value with a first preset threshold range and comparing the second feature value with a second preset threshold range, if both are within the respective preset threshold range, determining that the image quality is qualified, and executing step S13; if any one is not within the corresponding preset threshold range, determining it as abnormal data, and sending instructions to the image acquisition and processing module to trigger secondary acquisition and processing of the current target plate image;

[0070] S13: calling no less than three different image sharpness evaluation algorithms to perform parallel calculation on the image feature data qualified for quality, obtaining no less than three intermediate evaluation results;

[0071] S14: weighting and fusing the no less than three intermediate evaluation results, and matching the fusion result with a preset resolution lookup table to obtain a preliminary optical performance parameter value;

[0072] S15: comparing the preliminary optical performance parameter value with the historical parameter interval of the same type of instrument in the historical detection database, if the preliminary parameter value deviates from the historical parameter interval, marking it as data to be reviewed, and repeating steps S12 to S14 for review;

[0073] S16: outputting the data reviewed or confirmed as normal by step S15 as the verified optical performance parameter to the performance modeling and diagnosis unit.

[0074] It should be further explained that the image clarity evaluation and verification unit receives image feature data transmitted from the image acquisition and processing module, which at least includes the image edge gradient amplitude calculated by the Sobel edge detection algorithm as the first feature value, and the signal-to-noise ratio value obtained by calculating the ratio of the image signal standard deviation to the background noise standard deviation as the second feature value.

[0075] The unit first compares the first feature value with the preset gradient amplitude threshold range, and at the same time compares the second feature value with the preset signal-to-noise ratio qualified range; when both feature values are within their respective preset threshold ranges, it is determined that the current image quality meets the processing requirements, and the next step is entered; if any of the feature values does not meet the threshold requirement, it is determined that the frame of image data is abnormal, and the unit immediately sends a control instruction to the image acquisition and processing module to trigger a new round of image acquisition and feature extraction process for the same target plate target.

[0076] For image data that passes the initial quality screening, the unit calls an algorithm library containing Sobel operator, Canny operator and Laplace operator, and uses these three different image clarity evaluation algorithms to independently and in parallel operate on the same set of feature data, respectively generating three independent intermediate clarity evaluation values. The unit has a preset weight coefficient table corresponding to different algorithms, and according to the current ambient illuminance data, selects the corresponding weight combination to calculate the weighted sum of the above three intermediate evaluation values, and obtains a comprehensive clarity score.

[0077] Subsequently, the unit accesses its internally stored resolution lookup table, which establishes a mapping relationship between the comprehensive sharpness score and the standard resolution angle value, and obtains the preliminary optical performance parameter through matching. To ensure the reliability of the parameter, the unit further accesses the local historical database to retrieve the historical detection data interval of the same type of low-light night vision instrument under similar environmental conditions, and compares and analyzes the preliminary parameter with the historical data interval. If the parameter value deviates from the historical interval, the unit will mark this data as a pending review state and automatically repeat the complete verification process from image quality threshold comparison to historical data comparison. Only the data that successfully passes the historical data comparison or is confirmed as valid after the complete review process will be finally confirmed by the unit as the verified optical performance parameter and output to the subsequent performance modeling and diagnosis unit for in-depth analysis.

[0078] The pre-trained time series prediction model in the performance modeling and diagnosis unit is a long short-term memory neural network model, and its construction and operation include:

[0079] The input feature vector received by the input layer of the model includes: the verified optical performance parameter, the temperature data and humidity data collected by the environmental parameter acquisition module in real time, and the cumulative working time data of the measured low-light night vision instrument;

[0080] The hidden layer of the model contains multiple memory cells for learning and memorizing the change pattern and dependency relationship of the input feature vector in the time dimension;

[0081] The output of the output layer of the model includes: the optical performance parameter prediction value of the measured low-light night vision instrument in the next detection period, the health index for indicating the degree of performance degradation, and the time node information for suggesting maintenance operation.

[0082] It needs to be further explained that the pre-trained time series prediction model in the performance modeling and diagnosis unit is a long short-term memory neural network model, and its specific construction and operation process is as follows: the input layer of the model receives a feature vector composed of multi-dimensional data, which includes the optical performance parameter (such as resolution angle value) output by the image sharpness evaluation and verification unit after multiple rounds of verification, the temperature data and humidity data collected by the environmental parameter acquisition module in real time, and the instrument power-on working time data read from the control system of the measured low-light night vision instrument or recorded by the host.

[0083] The hidden layer of the model is composed of a series of neuron units with memory function, each of which contains a gating structure of input gate, forget gate and output gate inside; these structures work together to learn the dynamic change rule of the above-mentioned input feature vector at consecutive time steps, wherein the forget gate determines which information to retain from the memory state at the previous moment, the input gate controls which new information at the current moment to store in the memory state, and the output gate calculates the output of the unit based on the current memory state and input; through this mechanism, the model can capture the long-term dependence relationship and short-term fluctuation pattern of the optical performance parameters generated by the change of use time and environmental conditions.

[0084] The output layer of the model receives the processed information of the hidden layer, and after passing through a fully connected layer and an activation function, finally outputs a vector containing multiple specific information: the vector includes the predicted value of the optical performance parameter of the measured instrument at the next scheduled detection period calculated by the model, a quantitative health index that comprehensively reflects the deviation of the current performance state of the instrument from its initial state or health benchmark, and specific time node information of the next maintenance operation recommended to be performed based on the performance degradation trend and the preset maintenance threshold.

[0085] The performance modeling and diagnosis unit also includes a model updating module for performing iterative optimization of the model.

[0086] The model updating module stores the verified optical performance parameters obtained by each detection, corresponding environmental data and diagnosis results to the local historical database;

[0087] The model updating module periodically re-trains the long short-term memory neural network model using new data in the local historical database to update the internal weight parameters of the model.

[0088] The system supports data synchronization with a cloud server, and the cloud server aggregates historical data of multiple test systems of the same type to generate and distribute globally optimized model parameters to the local.

[0089] It needs to be further explained that the model updating module built-in the performance modeling and diagnosis unit is responsible for performing iterative optimization of the prediction model, and its workflow is as follows: after each detection task is completed, the module automatically packages all the relevant data obtained by the detection, including the verified optical performance parameters, the temperature and humidity data recorded by the corresponding environmental parameter acquisition module, and the diagnosis results and health index generated by the performance modeling and diagnosis unit, into a structured data record, adds a timestamp and a device identifier, and stores it in the solid state memory located in the local data processing and analysis host, forming a continuously growing local historical database.

[0090] The model updating module is configured with a settable time period trigger, for example, triggering the retraining process of the model after completing a set number of detections or after a fixed calendar time elapses; the process calls all or recent data in the local historical database, uses the back propagation algorithm common in the time series prediction field combined with the gradient descent optimizer to iteratively adjust the internal connection weight parameters of the long short-term memory neural network model, so that the output of the model is more consistent with the actual performance evolution trajectory of the specific instrument being tested or similar instruments.

[0091] To achieve a wider range of data-driven optimization, the system design supports establishing a secure connection with a remote cloud server through Ethernet or wireless network; the model updating module periodically uploads the anonymized detection data in the local historical database to the cloud server, the cloud server aggregates data uploaded from multiple test systems of the same type from various places, uses a larger data set for centralized model training, and generates a set of globally optimized model weight parameters; then, the cloud server distributes the updated parameters to each local system, and the model updating module receives the parameters and safely replaces the corresponding old parameters in the local model, so that the local performance prediction capability can continuously evolve based on more extensive instrument usage data.

[0092] The process of adjusting parameters by the data closed-loop feedback unit according to the performance state diagnosis result includes:

[0093] When the diagnosis result indicates that the performance of the micro-light night vision instrument being tested is in a stable state, the data closed-loop feedback unit keeps the current processing parameters of the image acquisition and processing module unchanged;

[0094] When the diagnosis result indicates that the performance of the micro-light night vision instrument being tested shows fluctuation or a downward trend, the data closed-loop feedback unit sends instructions to the image clarity evaluation and verification unit to adjust the algorithm weights used when performing weighted fusion, and increase the weight proportion of the algorithm that is more sensitive to image noise;

[0095] The data closed-loop feedback unit also dynamically adjusts the working gain of the image sensor in the image acquisition and processing module according to the illumination intensity information in the environmental data.

[0096] It needs to be further explained that the specific process of adjusting system parameters by the data closed-loop feedback unit according to the performance state diagnosis result is as follows: the unit continuously receives the diagnosis result output from the performance modeling and diagnosis unit, which contains the health index and performance trend identifier; when the health index in the diagnosis result continuously stays in the preset stable interval and the trend identifier shows "smooth", the unit determines that the performance state of the instrument being tested is stable, at this time, it does not send any parameter modification instructions to the image acquisition and processing module or the image clarity evaluation and verification unit, and maintains its existing working state.

[0097] When the health index in the diagnosis result reaches or is lower than the preset attention threshold, or the trend identification shows "decrease", the unit determines that the instrument performance may have fluctuation or degradation signs; at this time, the unit sends an adjustment instruction containing a specific parameter set to the image clarity evaluation and verification unit, which modifies the combination of weight coefficients used by the verification unit when performing multi-algorithm result weighted fusion, specifically by increasing the weight proportion of the Laplace operator evaluation result which is more sensitive to high-frequency noise in the image, and correspondingly reducing the weights of other operators, aiming to more sensitively capture the subtle degradation of image quality.

[0098] In addition, the data closed-loop feedback unit also receives the light intensity data stream sent by the environmental parameter acquisition module in real time; the unit has a pre-stored correspondence table of different light intensity intervals and image sensor gain suggestion values, and queries this table according to the current light intensity value and sends instructions to the image acquisition and processing module to dynamically set the analog gain or digital gain value of its image sensor, so as to ensure that the image signal intensity collected under different environmental illuminance remains within a range suitable for subsequent processing, thereby forming a double feedback mechanism of adaptive adjustment of front-end acquisition and processing parameters driven by both instrument internal performance diagnosis and external environmental conditions.

[0099] The image acquisition and processing module includes an image sensor and a signal processing circuit, the image sensor is optically coupled to the rear end of the eyepiece of the measured low-light night vision instrument, for directly capturing the optical image output by the eyepiece and converting it into a digital electrical signal; the signal processing circuit is used for preprocessing the digital electrical signal, including analog-to-digital conversion, noise filtering and illumination compensation, to generate image feature data.

[0100] It needs to be further explained that the image acquisition and processing module includes a high-sensitivity CMOS image sensor and a special signal processing circuit matched with it; the image sensor is directly installed behind the exit pupil position of the eyepiece of the low-light night vision instrument in a coaxial alignment manner through a customized adaptive interface with a locking mechanism, so that its photosensitive surface can completely receive and capture the optical image carrying the target plate pattern information emitted from the eyepiece, and convert the optical signal into an analog electrical signal; after receiving the analog electrical signal, the signal processing circuit first adjusts the signal amplitude through a programmable gain amplifier, and then converts it into digital image data through a high-precision analog-to-digital converter; the converted digital data is immediately sent to a field programmable gate array for pipeline processing, which includes applying a spatial filtering algorithm based on Gaussian kernel to suppress random noise, and performing an adaptive contrast stretching algorithm based on overall image gray histogram analysis to compensate for the influence of environmental light changes; after the above pre-processing, the circuit extracts a data packet containing specific image features, which encapsulates the calculated edge gradient information, local contrast value and regional signal-to-noise ratio value, and uploads it in real time to the data processing and analysis host through a serial communication interface.

[0101] The target plate module includes a support frame and a target plate detachably mounted on the top of the support frame, the front and back surfaces of the target plate are coated with different reflectivity diffuse reflection coatings, respectively forming high-contrast observation surface and low-contrast observation surface; the environmental parameter acquisition module is integrated on the support frame, including temperature sensor, humidity sensor and illuminance sensor.

[0102] It needs to be further explained that the target plate module includes a height-adjustable tripod as a support frame, a level bubble is installed on the tripod head of the tripod, which is used to keep the tripod head horizontal by adjusting the length of the support legs when setting up; a target plate made of homogeneous aluminum plate is detachably connected with the lower end of a vertical connecting rod through the threaded interface at the back center of the target plate, and the upper end of the connecting rod is fixed on the tripod head through a quick-release clamp seat, so as to realize the stable installation of the target plate on the top of the support frame.

[0103] The front surface of the target plate is coated with a white diffuse reflection coating with a reflectivity of about 85%, and the back surface is coated with a similar coating with a reflectivity of about 35%, so as to form different reflection characteristics on the front and back surfaces, respectively corresponding to high-contrast observation conditions and low-contrast observation conditions; the surface of the target plate is etched with a resolution test pattern conforming to the USAF-1951 standard.

[0104] The shell of the environmental parameter acquisition module is directly fixed on one of the support legs of the support frame, and the integrated sensors in the shell include a negative temperature coefficient thermistor for measuring the ambient air temperature, a capacitive humidity sensing element for measuring the relative humidity, and a silicon photodiode for measuring the illumination at the position of the target plate surface. The sensing parts of these sensors are exposed outside the shell and are connected to the data processing and analysis host through cables to realize real-time transmission of environmental data.

[0105] The data processing and analysis host also includes a man-machine interaction and display unit, which is used to:

[0106] real-time display of the original images collected by the image acquisition and processing module and the processed image feature data;

[0107] visually display the history curve of the verified optical performance parameters, the performance trend prediction curve generated by the performance modeling and diagnosis unit, and the health index in the form of a chart;

[0108] receive user input operation instructions and set system alarm thresholds.

[0109] It should be further noted that the man-machine interaction and display unit integrated in the data processing and analysis host includes a graphical user interface that runs on the display screen of the host and can be operated through touch or external input devices. The interface is divided into multiple functional areas: a main display area for real-time presentation of uncompressed original target plate images received from the image acquisition and processing module and contrast images after preprocessing and feature enhancement; a parameter monitoring area dynamically updates the optical performance parameters output from the image clarity evaluation and verification unit, as well as the temperature, humidity, and illumination values obtained from the environmental parameter acquisition module in the form of numbers and instrument panels; a historical trend area draws the historical curve of the optical performance parameters over time in the form of a two-dimensional line chart, while superimposing the performance prediction curve generated by the performance modeling and diagnosis unit extending to the future time point, and visually distinguishing the safety, attention, and warning levels of the instrument health index with different colored filled areas.

[0110] The unit also provides a configuration panel through which users can input or modify system parameters, such as setting specific numerical thresholds corresponding to different health index levels as system alarm conditions, or adjusting the initial weight coefficients of each algorithm in the image clarity evaluation and verification process. All operation instructions triggered by the user through the interface are analyzed and verified by the logic processing module inside the unit, and then converted into corresponding control commands and sent to other corresponding modules of the system for execution, while the interface state is updated in real time according to the system feedback.

[0111] The workflow of the system includes:

[0112] Step a: Set the target plate module at a preset detection distance from the micro-light night vision instrument to be tested, start the instrument and align it to the target plate module;

[0113] Step b: The image acquisition and processing module acquires images and performs preprocessing, while the environmental parameter acquisition module acquires environmental data, both of which send data to the data processing and analysis host;

[0114] Step c: The image clarity evaluation and verification unit performs multiple rounds of parallel calculation and cross-verification on the received data, and outputs the verified optical performance parameters after rechecking and confirmation;

[0115] Step d: The performance modeling and diagnosis unit combines environmental data and uses a time series prediction model to analyze the verified optical performance parameters, generating a diagnosis result containing prediction information;

[0116] Step e: The data closed-loop feedback unit adjusts the relevant processing parameters in the subsequent detection of the system according to the diagnosis result.

[0117] It needs to be further explained that the system performs detection according to the following workflow: First, the operator determines a preset detection distance according to the specification parameters of the micro-light night vision instrument to be tested, sets up the target plate module at the distance, and uses a laser ranging device to recheck the distance; start the instrument and adjust its orientation to make its optical axis aligned with the center of the divided pattern on the target plate module.

[0118] Subsequently, the system starts, the image acquisition and processing module begins to continuously acquire images output from the objective lens of the instrument to be tested, synchronously performs preprocessing including noise filtering and contrast enhancement, and extracts image feature data; the environmental parameter acquisition module simultaneously begins to acquire temperature, humidity and illumination data at the target plate; these two types of data are transmitted to the data processing and analysis host in real time.

[0119] Next, the image clarity evaluation and verification unit of the host performs a multi-step verification process including initial quality screening, multiple algorithm parallel operation, result weighted fusion and historical data interval comparison on the received image feature data; this process ensures that the output optical performance parameters have undergone not less than two independent calculations and consistency checks.

[0120] Then, the performance modeling and diagnosis unit inputs the verified optical performance parameters, environmental data and instrument working time data into the pre-trained long short-term memory neural network model; the model analyzes the time dependence of the parameter sequence and outputs a comprehensive diagnosis result containing future performance prediction value, quantitative health index and maintenance time node.

[0121] Finally, the data closed-loop feedback unit generates a parameter adjustment instruction according to the performance state trend reflected in the diagnosis result; the instruction is sent to the image acquisition and processing module or the image sharpness evaluation and verification unit to dynamically optimize the processing gain or algorithm weight parameter in the subsequent detection, thereby completing a closed-loop test cycle from data acquisition, intelligent verification, modeling diagnosis to parameter self-adaptive adjustment.

[0122] In step c, the number of multiple rounds of parallel calculation and cross verification is not less than two, and each verification is based on an independent algorithm combination or parameter setting; after step e, the system archives the complete data package of this detection, including the original data, intermediate processing data, final result and adjusted parameters, to the local historical database for subsequent model updating and analysis tracing.

[0123] It needs to be further explained that in the detection process, the multiple rounds of parallel calculation and cross verification performed by the image sharpness evaluation and verification unit are set to be at least two complete processing cycles; in each independent verification cycle, the unit calls a set of preset algorithm combinations, which at least includes three different sharpness evaluation algorithms, and the algorithm types or internal parameter settings used in adjacent two verification cycles are independent of each other, for example, the first cycle uses Sobel, Canny and Tenengrad algorithm based on gradient, and the second cycle uses Laplacian operator, variance-based method and frequency spectrum-based method, so as to ensure that the evaluation result is not dependent on a single algorithm or fixed parameter set.

[0124] After each verification cycle is completed, the unit not only outputs the final confirmed optical performance parameter, but also encapsulates the complete intermediate data generated in this cycle, including the original input image feature value, the intermediate calculation result of each independent algorithm, the weight fusion process data and the historical data comparison log, together with the environmental data, instrument identifier and time stamp of this detection.

[0125] After receiving the instruction to complete all verification processes, the data processing and analysis host immediately calls its data management module to write the above-mentioned encapsulated complete data package in a structured format (such as JSON or custom binary format) to a special historical database partition in the local non-volatile memory; this archiving operation ensures that all related information from the original input to the final conclusion of each detection is persistently saved and indexed.

[0126] These archived data provide detailed training and verification data sets for the model updating module in the subsequent performance modeling and diagnosis unit, at the same time, the complete data chain also supports the backtracking analysis of any historical detection result, when performance anomaly occurs, the corresponding data package can be called to review all processing intermediate links, thereby realizing the full recording and traceability of the detection process.

[0127] The test system replaces the traditional method of relying on the visual judgment of operating personnel by automatic image acquisition and processing, combined with multi-algorithm parallel computing and cross-verification processes, effectively reducing the influence of subjective factors and environmental fluctuations on the detection results, and improving the consistency and objectivity of the detection results. The system can automatically complete the whole process from image acquisition, feature extraction to parameter verification, improving the efficiency and reliability of the detection operation.

[0128] In addition, the system uses a time series prediction model to fuse and analyze the performance parameters and environmental data after verification, realizes quantitative evaluation and prediction of the instrument performance degradation trend, and forms a closed-loop feedback based on the diagnosis results to dynamically adjust the front-end processing parameters. This not only extends the performance evaluation from single detection to long-term state monitoring, but also provides data support for preventive maintenance, enhancing the adaptive ability and intelligent level of the test system.

[0129] It should be noted that, in this article, relational terms such as first and second are used merely to distinguish one entity or action from another, without necessarily requiring or implying any such actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof are intended to cover non-exclusive inclusions, so that a process, method, article, or apparatus including a series of elements includes not only those elements, but also other elements not explicitly listed, or inherent to such a process, method, article, or apparatus. Without more limitations, the element defined by the statement "comprises a" does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0130] Although embodiments of the present application have been shown and described, it will be understood by those having ordinary skill in the art that various changes, modifications, substitutions and alterations can be made hereto without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A low-light night vision optical performance testing system based on data modeling, characterized in that, include: The target plate module is used to provide the low-light night vision instrument under test with an observation target featuring a reticle pattern. The image acquisition and processing module is used to acquire the image output by the low-light night vision instrument under test after observing the target plate module, and to process the image to extract image feature data; An environmental parameter acquisition module is used to collect environmental data around the target module in real time. The data processing and analysis host is connected to the image acquisition and processing module and the environmental parameter acquisition module, respectively. The data processing and analysis host includes: The image sharpness assessment and verification unit is used to receive the image feature data, perform parallel calculations and cross-verifications on the image feature data based on a variety of preset image sharpness assessment algorithms, and output the verified optical performance parameters. The performance modeling and diagnosis unit is used to receive the verified optical performance parameters and the environmental data, analyze them based on the pre-trained time series prediction model, and generate the performance status diagnosis results and trend prediction information of the tested low-light night vision instrument. The data closed-loop feedback unit is used to receive the diagnostic results output by the performance modeling and diagnostic unit, and dynamically adjust the processing parameters of the image acquisition and processing module or the image sharpness assessment and verification unit according to the diagnostic results.

2. The low-light night vision optical performance testing system based on data modeling according to claim 1, characterized in that: The image sharpness assessment and verification unit performs the following steps for verification: S11: Receive the image feature data from the image acquisition and processing module, wherein the image feature data includes at least a first feature value calculated based on the edge gradient algorithm and a second feature value calculated based on the image signal-to-noise ratio algorithm; S12: Compare the first feature value with the first preset threshold range, and compare the second feature value with the second preset threshold range. If both are within their respective preset threshold ranges, the image quality is deemed acceptable, and step S13 is executed. If either is not within its corresponding preset threshold range, it is deemed abnormal data, and an instruction is sent to the image acquisition and processing module to trigger secondary acquisition and processing of the current target image. S13: Call no less than three different image sharpness evaluation algorithms to perform parallel calculations on the image feature data that meets the quality requirements, and obtain no less than three intermediate evaluation results; S14: The not less than three intermediate evaluation results are weighted and fused, and the fusion result is matched with a preset resolution lookup table to obtain preliminary optical performance parameter values. S15: Compare the preliminary optical performance parameter values ​​with the historical parameter range of the same model of instrument in the historical test database. If the preliminary parameter values ​​deviate from the historical parameter range, they are marked as data to be reviewed, and steps S12 to S14 are repeated for review. S16: Data that has been verified or confirmed as normal in step S15 is output to the performance modeling and diagnosis unit as the verified optical performance parameters.

3. The low-light night vision optical performance testing system based on data modeling according to claim 2, characterized in that: The time series prediction model pre-trained in the performance modeling and diagnosis unit is a long short-term memory neural network model, and its construction and operation include: The input feature vector received by the input layer of the model includes: the verified optical performance parameters, the temperature and humidity data collected in real time by the environmental parameter acquisition module, and the cumulative working time data of the tested low-light night vision instrument. The hidden layer of the model contains multiple memory units, which are used to learn and memorize the change patterns and dependencies of the input feature vector in the time dimension. The output layer of the model includes: predicted optical performance parameters of the tested low-light night vision instrument in the next testing cycle, a health index indicating the degree of performance degradation, and time node information for recommending maintenance operations.

4. The low-light night vision optical performance testing system based on data modeling according to claim 3, characterized in that: The performance modeling and diagnostic unit also includes a model update module for performing iterative optimization of the model: The model update module stores the verified optical performance parameters, the corresponding environmental data, and the diagnostic results obtained from each detection into a local historical database. The model update module periodically uses new data from the local historical database to retrain the long short-term memory neural network model in order to update the model's internal weight parameters. The system supports data synchronization with a cloud server, which aggregates historical data from multiple test systems of the same model to generate and distribute globally optimized model parameters to the local system.

5. The low-light night vision optical performance testing system based on data modeling according to claim 4, characterized in that: The process by which the data closed-loop feedback unit adjusts parameters based on the performance status diagnosis results includes: When the diagnostic result indicates that the performance of the tested low-light night vision instrument is in a stable state, the data closed-loop feedback unit keeps the current processing parameters of the image acquisition and processing module unchanged; When the diagnostic results indicate that the performance of the tested low-light night vision instrument is fluctuating or declining, the data closed-loop feedback unit sends an instruction to the image clarity assessment and verification unit to adjust the algorithm weights used in the weighted fusion and increase the weight ratio of algorithms that are more sensitive to image noise. The data closed-loop feedback unit also dynamically adjusts the working gain of the image sensor in the image acquisition and processing module based on the light intensity information in the environmental data.

6. The low-light night vision optical performance testing system based on data modeling according to claim 5, characterized in that: The image acquisition and processing module includes an image sensor and a signal processing circuit. The image sensor is optically coupled to the rear end of the eyepiece of the low-light night vision instrument under test, and is used to directly capture the optical image output by the eyepiece and convert it into a digital electrical signal. The signal processing circuit is used to perform preprocessing on the digital electrical signal, including analog-to-digital conversion, noise filtering and illumination compensation, to generate the image feature data.

7. The low-light night vision optical performance testing system based on data modeling according to claim 6, characterized in that: The target plate module includes a support frame and a target plate that can be detachably installed on the top of the support frame. The front and back sides of the target plate are coated with diffuse reflection coatings with different reflectivities to form a high-contrast observation surface and a low-contrast observation surface, respectively. The environmental parameter acquisition module is integrated on the support frame and includes a temperature sensor, a humidity sensor and an illuminance sensor.

8. The low-light night vision optical performance testing system based on data modeling according to claim 7, characterized in that: The data processing and analysis host also includes a human-computer interaction and display unit, which is used for: The image acquisition and processing module can display the original image and the processed image feature data in real time. The historical change curves of the verified optical performance parameters, the performance trend prediction curves generated by the performance modeling and diagnosis unit, and the health index are visualized in chart form. Receive user input commands and set system alarm thresholds.

9. A low-light night vision optical performance testing system based on data modeling according to claim 8, characterized in that: The system's workflow includes: Step a: Set the target plate module and the tested low-light night vision instrument to a preset detection distance, start the tested low-light night vision instrument and align it with the target plate module; Step b: The image acquisition and processing module acquires images and performs preprocessing, while the environmental parameter acquisition module acquires environmental data. Both modules send the data to the data processing and analysis host. Step c: The image sharpness evaluation and verification unit performs multiple rounds of parallel computation and cross-verification on the received data, and outputs the verified optical performance parameters after review and confirmation; Step d: The performance modeling and diagnosis unit combines the environmental data and uses the time series prediction model to analyze the verified optical performance parameters, and generates the diagnosis result containing prediction information; Step e: The data closed-loop feedback unit adaptively adjusts the relevant processing parameters in subsequent system detection based on the diagnostic results.

10. A low-light night vision optical performance testing system based on data modeling according to claim 9, characterized in that: In step c, the number of rounds of parallel computation and cross-verification is no less than two, and each verification is based on an independent combination of algorithms or parameter settings. After step e, the system archives and saves the complete data packet of this detection, including the original data, intermediate processing data, final results and adjusted parameters, to the local historical database for subsequent model updates and analysis traceability.

Images