Construction method and system of RGB intelligent night vision light source

Abstract

The invention relates to the technical field of intelligent night vision light source construction, and discloses an RGB intelligent night vision light source construction method and system, and the method comprises the steps: collecting the environment light intensity and other data of a light source construction region, obtaining environment perception data, constructing an environment feature tensor, and obtaining an initial control parameter through a dual-delay network; inquiring an LED array, configuring spatial light and pulse width, and obtaining a tunable composite spectrum by using initial parameter driving; a light reflection target is inquired, spectrum irradiation is used, data are collected, and an illumination-imaging joint optimization model is constructed through processing to obtain target data; target features are identified by using target data, a scene semantic vector and a light source control strategy are constructed, unit parameters are optimized through matrix calculation and decomposition, equipment is deployed to obtain a target LED array, and the RGB intelligent night vision light source is constructed. According to the invention, the accuracy of night vision imaging of the RGB intelligent night vision light source can be improved.

Description

Technical Field

[0001] The invention relates to a method and system for constructing an RGB intelligent night vision light source, and belongs to the technical field of intelligent night vision light source construction. Background Art

[0002] Night vision light source technology has important application value in security monitoring, military reconnaissance, autonomous driving and medical imaging. Especially in low-light or complex lighting environments, its performance directly determines the clarity and reliability of the imaging system. Traditional night vision light sources mostly rely on single-wavelength infrared LEDs or fixed-spectrum fill-in solutions to enhance the light intensity of specific bands to improve imaging effects. However, this type of method has significant limitations in practical applications: first, fixed-spectrum light sources are difficult to adapt to changing scene requirements (such as haze, reflections, dynamic targets, etc.), resulting in unstable imaging quality; second, there is a lack of intelligent control capabilities, and it is impossible to adjust the light source parameters in real time according to factors such as ambient light intensity, target reflection characteristics and noise interference. It needs to rely on manual intervention, which is inefficient and easy to introduce human errors; in addition, traditional solutions have deficiencies in the coordinated use of visible light and near-infrared spectra, making it difficult to achieve a balance between high-fidelity color reproduction and detail enhancement, limiting the comprehensive performance of night vision systems.

[0003] At present, some improved technologies attempt to expand the adaptability of light sources through multi-band LED arrays or filter switching mechanisms, but such methods usually rely on preset mode switching, have slow response speeds and high power consumption, and are difficult to meet the needs of real-time dynamic scenes. At the same time, existing systems mostly use discrete control strategies, and the light source parameters (such as brightness, color temperature, and band share) lack deep coordination with the feedback of imaging sensors, resulting in low resource utilization and limited noise suppression capabilities. For example, under complex background interference, a fixed fill light mode may over-enhance irrelevant areas and instead cover up key target information. Therefore, there is an urgent need for a method for constructing a night vision light source that can deeply integrate RGB multi-spectral intelligent control, environmental perception, and real-time feedback to improve the accuracy of night vision imaging. Summary of the invention

[0004] The present invention provides a method and system for constructing an RGB intelligent night vision light source, the main purpose of which is to improve the accuracy of night vision imaging of the RGB intelligent night vision light source.

[0005] To achieve the above object, the present invention provides a method for constructing an RGB intelligent night vision light source, comprising:

[0006] Collecting ambient light intensity, spectral distribution and dynamic target thermal radiation data of the light source construction area corresponding to the RGB intelligent night vision light source to obtain environmental perception data, constructing an environmental feature tensor of the environmental perception data, and based on the environmental feature tensor, using a double delay network to construct initial control parameters of the RGB intelligent night vision light source;

[0007] Querying the LED array of the RGB intelligent night vision light source, configuring the spatial light and pulse width of the LED array, obtaining a preliminary configured LED array, and driving the preliminary configured LED array using the initial control parameters to obtain a tunable composite spectrum of the RGB intelligent night vision light source;

[0008] Query the light reflection target of the light source construction area, illuminate the light reflection target with the tunable composite spectrum, collect light reflection data of the light reflection target, identify the data sparsity of the light reflection data, configure the observation matrix of the light reflection data based on the data sparsity, construct high dynamic range spectral cube data of the light reflection data with the observation matrix, construct an illumination-imaging joint optimization model of the light source construction area, optimize the high dynamic range spectral cube data with the illumination-imaging joint optimization model, and obtain target data;

[0009] The target data is used to identify the surface material characteristics and micro-motion features of the light reflective target, and the surface material characteristics and the micro-motion features are used to construct a scene semantic vector of the light source construction area. The scene semantic vector is combined with a preconfigured strategy network to construct a light source control strategy for the preliminary configured LED array, and a frequency domain-spatial domain joint optimization matrix of the light source control strategy is constructed. The frequency domain-spatial domain joint optimization matrix is ​​non-negative tensor decomposition is performed on the frequency domain-spatial domain joint optimization matrix to obtain optimization unit parameters for the preliminary configured LED array, and equipment deployment is performed on the preliminary configured LED array based on the optimization unit parameters to obtain a target LED array, and the target LED array is used to construct an RGB intelligent night vision light source for the light source construction area.

[0010] Optionally, the initial control parameters of the RGB intelligent night vision light source are constructed using a dual delay network based on the environmental feature tensor, including:

[0011] Deploy a dual-delay network of the RGB intelligent night vision light source;

[0012] Using the environmental feature tensor to construct a policy network and a value network of the dual-delay network to obtain a deployed dual-delay network;

[0013] Initializing network parameters of the deployed double-delay network, and constructing an experience playback buffer of the deployed double-delay network to obtain a target double-delay network;

[0014] After inputting the environment feature tensor into the policy network of the target dual-delay network, sampling buffer data from the experience replay buffer to obtain sampled data;

[0015] Using the sampled data, updating the value network in the target double-delay network to obtain an updated double-delay network;

[0016] After the environmental feature tensor is input into the updated double-delay network, the output data of the updated double-delay network is collected to obtain initial control parameters.

[0017] Optionally, the using the sampled data to perform a network update on the value network in the target double-delay network to obtain an updated double-delay network includes:

[0018] Extracting sampling actions and sampling states from the sampling data;

[0019] After the sampling action and the sampling state are input into the value network, the target value of the value network is calculated using the following formula:

[0020] y i =r i +γmin j＝1，2 Q′(s i+1 ,π′(s i+1 );θ′ j )

[0021] Among them, y i represents the target value, r i represents the i-th reward value in the sampled data, γ represents the discount factor of the value network, Q′ represents the target value network corresponding to the value network, and s i+1 represents the i+1th sampling state, θ′ j represents the jth parameter of the target value network;

[0022] Based on the target value, the mean square error of the value network is calculated using the following formula:

[0023]

[0024] Among them, L(θ) represents the mean square error, N represents the number of sampled data, and y i represents the target value, s i represents the i-th sampling state, a i represents the i-th sampled action, θ represents the parameters of the value network;

[0025] Based on the mean square error, the value network in the target double-delay network is updated by using a gradient descent method to obtain an updated double-delay network.

[0026] Optionally, configuring the spatial light and pulse width of the LED array to obtain a preliminary configuration of the LED array includes:

[0027] identifying a visual scene of the LED array;

[0028] Based on the visual scene, determining a target spatial light distribution of the LED array;

[0029] Performing illumination simulation on the LED array according to the target spatial light distribution, and extracting preferred device parameters corresponding to preferred simulation results in the illumination simulation;

[0030] configuring the spatial light of the LED array using the preferred device parameters;

[0031] querying a visual scene of the LED array and identifying target characteristics in the visual scene;

[0032] Based on the target characteristics, construct a pulse width configuration strategy for the LED array;

[0033] After the pulse width of the LED array is configured using the pulse width configuration strategy, a preliminary configured LED array is obtained.

[0034] Optionally, identifying data sparsity of the light reflection data includes:

[0035] Converting the light reflection data into vector data;

[0036] Arrange elements of the vectors in the vector data to obtain arrangement elements;

[0037] The Gini coefficient of the arrangement element is calculated using the following formula:

[0038]

[0039] Among them, β represents the Gini coefficient, n represents the number of permutation elements, represents the absolute mean of the array elements, i represents the search variable of the array elements, x i Represents the i-th element in the array;

[0040] Based on the Gini coefficient, the data sparsity of the light reflection data is analyzed.

[0041] Optionally, constructing high dynamic range spectral cube data of the light reflection data using the observation matrix includes:

[0042] Performing data preprocessing on the light reflection data to obtain preprocessed data;

[0043] Projecting the preprocessed data into the measurement matrix to obtain projection data;

[0044] identifying dimensional information of the projection data;

[0045] constructing a spectral cube architecture of the projection data based on the dimensional information;

[0046] Using the projection data to fill the spectral cube architecture with data to obtain a filled spectral cube;

[0047] Reconstructing the light reflection data to obtain reconstructed data;

[0048] Filling the reconstructed data into the filled spectral cube to obtain spectral cube data;

[0049] The spectral cube data is processed a posteriori to obtain high dynamic range spectral cube data.

[0050] Optionally, constructing the illumination-imaging joint optimization model of the light source construction area includes:

[0051] Query the lighting-imaging requirements of the light source construction area;

[0052] Collecting system data of the light source construction area based on the lighting-imaging requirements;

[0053] Identifying light source characteristics of the light source construction area;

[0054] Based on the light source characteristics, construct an illumination-imaging model framework of the light source construction area;

[0055] Using the lighting-imaging requirements, constructing an optimization objective function of the lighting-imaging model architecture;

[0056] Analyzing the physical limitations of the illumination-imaging device in the light source construction area to construct the constraints of the illumination-imaging model architecture;

[0057] Based on the optimization objective function and the constraint conditions, the illumination-imaging model architecture is deployed to obtain an illumination-imaging model;

[0058] Optimizing the illumination-imaging model using a genetic algorithm to obtain an optimized model;

[0059] Collecting real-time environmental data of the light source construction area to perform model verification on the optimization model and obtain verification parameters;

[0060] An index evaluation is performed on the verification parameters, and when an evaluation result of the index evaluation meets a preset result, an illumination-imaging joint optimization model is obtained.

[0061] Optionally, the identifying the surface material characteristics and micro-motion characteristics of the light reflective target by using the target data includes:

[0062] Extracting spectral data about the light reflective target from the target data;

[0063] Analyzing the spectral characteristics of the light reflective target at different wavelengths using the spectral data;

[0064] Collecting image information of the light reflection target;

[0065] Constructing a gray-level co-occurrence matrix of the light-reflecting target;

[0066] Analyzing texture information of the light reflective target using the gray level co-occurrence matrix based on the image information;

[0067] Identifying the surface material characteristics of the light reflective target using the spectral characteristics and the texture information;

[0068] Collecting an inter-frame difference image of the light reflective target;

[0069] Calculating the optical flow field of the inter-frame difference image;

[0070] The optical flow field is used to identify micro-motion features of the optical reflective target.

[0071] Optionally, constructing a frequency domain-spatial domain joint optimization matrix of the light source control strategy includes:

[0072] Query the frequency domain range and spatial domain range of the light source construction area corresponding to the light source control strategy;

[0073] Constructing a frequency domain matrix of the light source construction area based on the frequency domain range;

[0074] Constructing a spatial matrix of the light source construction area based on the spatial range;

[0075] Using the frequency domain matrix and the spatial domain matrix, construct a frequency domain-spatial domain joint matrix of the light source control strategy;

[0076] Filling the frequency-domain-spatial-domain joint matrix with elements using matrix information of the frequency-domain matrix and the spatial-domain matrix to obtain a filled matrix;

[0077] defining a matrix optimization function for the filling matrix;

[0078] The matrix optimization function is used to perform matrix optimization on the filling matrix to obtain a frequency domain-spatial domain joint optimization matrix.

[0079] In order to solve the above problems, the present invention also provides a system for constructing an RGB intelligent night vision light source, the system comprising:

[0080] An initial parameter construction module is used to collect ambient light intensity, spectral distribution and dynamic target thermal radiation data of the light source construction area corresponding to the RGB intelligent night vision light source, obtain environmental perception data, construct an environmental feature tensor of the environmental perception data, and construct the initial control parameters of the RGB intelligent night vision light source using a double delay network based on the environmental feature tensor;

[0081] A device parameter configuration module, used for querying the LED array of the RGB intelligent night vision light source, configuring the spatial light and pulse width of the LED array, obtaining a preliminary configured LED array, and driving the preliminary configured LED array using the initial control parameters to obtain a tunable composite spectrum of the RGB intelligent night vision light source;

[0082] An environmental data optimization module is used to query the light reflection target in the light source construction area, illuminate the light reflection target with the tunable composite spectrum, collect light reflection data of the light reflection target, identify the data sparsity of the light reflection data, configure the observation matrix of the light reflection data based on the data sparsity, construct high dynamic range spectral cube data of the light reflection data with the observation matrix, construct an illumination-imaging joint optimization model for the light source construction area, and optimize the high dynamic range spectral cube data with the illumination-imaging joint optimization model to obtain target data;

[0083] The intelligent night vision light source construction module is used to use the target data to identify the surface material characteristics and micro-motion features of the light reflective target, use the surface material characteristics and the micro-motion features to construct a scene semantic vector of the light source construction area, use the scene semantic vector in combination with a preconfigured strategy network to construct a light source control strategy for the preliminary configured LED array, construct a frequency domain-spatial domain joint optimization matrix for the light source control strategy, perform non-negative tensor decomposition on the frequency domain-spatial domain joint optimization matrix to obtain optimization unit parameters for the preliminary configured LED array, perform equipment deployment on the preliminary configured LED array based on the optimization unit parameters to obtain a target LED array, and use the target LED array to construct an RGB intelligent night vision light source for the light source construction area.

[0084] Compared with the problems described in the background technology, the embodiments of the present invention obtain environmental perception data by collecting ambient light intensity, spectral distribution and dynamic target thermal radiation data in the area where the light source is constructed, comprehensively understand the environmental conditions and target characteristics, improve the accuracy of target detection and recognition, and use a double-delay network to construct initial control parameters, so that the light source has good lighting and target detection performance when it starts working, reduce blind adjustments, and improve system efficiency; further, the embodiments of the present invention use the initial control parameters to drive the preliminary configuration of the LED array to obtain a tunable composite spectrum, thereby achieving preliminary adjustment of the spectral characteristics of the light source, enhancing the contrast between the target and the background, and improving the target detection and recognition capabilities; further, the embodiments of the present invention improve data collection efficiency by accurately querying the light reflection target and adjusting the spectral illumination parameters in a targeted manner. rate and quality, and then construct high dynamic range spectral cube data to facilitate comprehensive analysis of the light reflection target spectrum and spatial distribution, and then coordinate the relationship between lighting and imaging by constructing a lighting-imaging joint optimization model, give full play to the performance advantages of the light source and imaging system, improve the target imaging quality, reduce noise interference, provide a reliable basis for data processing, and improve system efficiency; further, the present invention finds the optimal combination of light sources by constructing a frequency domain-spatial domain joint optimization matrix to improve the light source performance, and obtains the target LED array based on the optimized parameter allocation equipment to ensure that the LED array works according to the optimization strategy, achieves the best performance of the light source, improves the quality and efficiency of the lighting system, and uses the target LED array to construct an RGB intelligent night vision light source, which can meet night vision needs, provide customized lighting solutions, and improve night observation and monitoring accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] Figure 1 A schematic diagram of a process for constructing an RGB intelligent night vision light source provided by an embodiment of the present invention;

[0086] Figure 2 A schematic diagram of a module for implementing a method for constructing an RGB intelligent night vision light source provided in one embodiment of the present invention.

[0087] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0088] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0089] The embodiment of the present application provides a method for constructing an RGB intelligent night vision light source. The execution subject of the method for constructing an RGB intelligent night vision light source includes but is not limited to at least one of the electronic devices such as a server and a terminal that can be configured to execute the method provided by the embodiment of the present application. In other words, the method for constructing an RGB intelligent night vision light source can be executed by software or hardware installed on a terminal device or a server device. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.

[0090] Embodiment 1:

[0091] Reference Figure 1 FIG. 1 is a flow chart of a method for constructing an RGB intelligent night vision light source provided by an embodiment of the present invention. In this embodiment, the method for constructing an RGB intelligent night vision light source includes:

[0092] S1. Collect the ambient light intensity, spectral distribution and dynamic target thermal radiation data of the light source construction area corresponding to the RGB intelligent night vision light source to obtain environmental perception data, construct an environmental feature tensor of the environmental perception data, and based on the environmental feature tensor, use a dual-delay network to construct the initial control parameters of the RGB intelligent night vision light source.

[0093] The embodiment of the present invention obtains environmental perception data by collecting the ambient light intensity, spectral distribution and dynamic target thermal radiation data of the light source construction area corresponding to the RGB intelligent night vision light source, so as to fully understand the environmental conditions and target characteristics of the light source construction area, so as to improve the accuracy of target detection and recognition.

[0094] Among them, the ambient light intensity refers to the intensity of the ambient light in the light source construction area corresponding to the RGB intelligent night vision light source, the spectral distribution refers to the energy distribution of light of different wavelengths in the ambient light in the light source construction area corresponding to the RGB intelligent night vision light source, and the dynamic target thermal radiation data refers to the electromagnetic waves emitted by objects in motion within the light source construction area, such as moving people and vehicles, due to their own temperature.

[0095] Optionally, the ambient light intensity can be collected by a light intensity sensor, the spectral distribution can be obtained by analyzing with a spectrometer, and the dynamic target thermal radiation data can be acquired by a thermal radiation detector.

[0096] The embodiment of the present invention can retain the spatial and feature information of the data by constructing the environmental feature tensor of the environmental perception data, which helps to extract richer and more accurate environmental features.

[0097] Optionally, the environmental feature tensor can be obtained by normalizing the collected environmental perception data such as ambient light intensity, spectral distribution and dynamic target thermal radiation, and organizing them into a multidimensional array form according to certain rules after preprocessing.

[0098] The embodiment of the present invention constructs the initial control parameters of the RGB intelligent night vision light source based on the environmental feature tensor using a dual delay network, so that the light source has better lighting effect and target detection performance when it starts working, and provides a good starting point for subsequent optimization and adjustment, reducing the process of blind trial and adjustment, and improving the efficiency and performance of the system.

[0099] The double-delay network refers to an algorithm and network architecture in deep reinforcement learning, which is often used in reinforcement learning tasks in continuous action space.

[0100] As an embodiment of the present invention, the initial control parameters of the RGB intelligent night vision light source are constructed using a dual-delay network based on the environmental feature tensor, including: deploying the dual-delay network of the RGB intelligent night vision light source, constructing a strategy network and a value network of the dual-delay network using the environmental feature tensor to obtain a deployed dual-delay network, initializing the network parameters of the deployed dual-delay network, and constructing an experience playback buffer of the deployed dual-delay network to obtain a target dual-delay network, inputting the environmental feature tensor into the strategy network of the target dual-delay network, sampling buffered data from the experience playback buffer to obtain sampled data, using the sampled data to perform network update on the value network in the target dual-delay network to obtain an updated dual-delay network, and inputting the environmental feature tensor into the updated dual-delay network, collecting the output data of the updated dual-delay network to obtain initial control parameters.

[0101] Among them, the policy network refers to a network model used to generate the actions of an intelligent agent (such as a system for controlling an RGB intelligent night vision light source), the value network refers to a network model that evaluates the value of an action taken by an intelligent agent under a specific environmental state, and the experience playback buffer refers to a data structure used to store and reuse the experience of the intelligent agent interacting with the environment. In the process of interacting with the environment, the intelligent agent will generate a series of experience data, usually expressed in the form of a tuple, i.e. (state, action, reward, next state). For example, in the control of an RGB intelligent night vision light source, each time the light source parameters (action) are adjusted according to the output of the policy network, a new environmental state (next state) will be observed and a corresponding reward (such as a reward for improved target visibility) will be obtained. This information constitutes an experience tuple.

[0102] Optionally, the dual-delay network can be deployed through a deep reinforcement learning model. The deployed dual-delay network can be based on the dual-delay network architecture, using structures such as multi-layer perceptrons, taking the environmental feature tensor as the input basis, and respectively building a strategy network for outputting light source control actions and a value network for evaluating action values, and forming a combination. The network parameters of the deployed dual-delay network can be initialized using a random initialization method, and the experience replay buffer can be constructed using a relational database (such as SQLite). The sampling data can input the environmental feature tensor into the strategy network to output an action, and then randomly extract a certain number of experience data tuples (state, action, reward, next state) from the experience replay buffer.

[0103] Further, as an optional embodiment of the present invention, the use of the sampling data to update the value network in the target dual-delay network to obtain an updated dual-delay network includes: extracting the sampling action and sampling state in the sampling data, inputting the sampling action and the sampling state into the value network, and calculating the target value of the value network using the following formula:

[0104] y i =r i +γmin j＝1，2 Q′(s i+1 ,π′(s i+1 );θ′ j )

[0105] Among them, y i represents the target value, r i represents the i-th reward value in the sampled data, γ represents the discount factor of the value network, Q′ represents the target value network corresponding to the value network, and s i+1 represents the i+1th sampling state, θ′ j represents the jth parameter of the target value network;

[0106] Based on the target value, the mean square error of the value network is calculated using the following formula:

[0107]

[0108] Among them, L(θ) represents the mean square error, N represents the number of sampled data, and y i represents the target value, s i represents the i-th sampling state, a i represents the i-th sampled action, θ represents the parameters of the value network;

[0109] Based on the mean square error, the value network in the target double-delay network is updated by using a gradient descent method to obtain an updated double-delay network.

[0110] S2. Query the LED array of the RGB intelligent night vision light source, configure the spatial light and pulse width of the LED array, obtain a preliminarily configured LED array, drive the preliminarily configured LED array using the initial control parameters, and obtain a tunable composite spectrum of the RGB intelligent night vision light source.

[0111] The embodiment of the present invention can collect and confirm comprehensive information about the LED array used by the RGB intelligent night vision light source by querying the LED array of the RGB intelligent night vision light source. This includes understanding the basic parameters of the LED array, such as the number of LED units, the layout method, and the luminous characteristics of each LED (such as the central wavelength, half-height width, etc.), as well as its electrical connection method, etc., thereby providing basic data for subsequent configuration work.

[0112] Among them, the LED array is a lighting or light-emitting component composed of multiple light-emitting diodes (LEDs) arranged and combined according to certain rules, and is widely used in fields such as RGB intelligent night vision light sources.

[0113] Furthermore, the embodiment of the present invention obtains a preliminary configuration of the LED array by configuring the spatial light and pulse width of the LED array so that the light source can better adapt to different application scenarios and target detection requirements and optimize energy utilization efficiency.

[0114] As an embodiment of the present invention, the spatial light and pulse width of the LED array are configured to obtain a preliminary configuration of the LED array, including: identifying a visual scene of the LED array, determining a target spatial light distribution of the LED array based on the visual scene, performing a lighting simulation on the LED array according to the target spatial light distribution, and extracting preferred device parameters corresponding to the preferred simulation results in the lighting simulation, configuring the spatial light of the LED array using the preferred device parameters, querying the visual scene of the LED array, and identifying target characteristics in the visual scene, constructing a pulse width configuration strategy for the LED array based on the target characteristics, and obtaining a preliminary configuration of the LED array after performing pulse width configuration on the LED array using the pulse width configuration strategy.

[0115] Optionally, the visual scene can be identified by collecting information such as images and light intensity of the LED array application scene through image sensors, environmental sensors, etc. The target spatial light distribution can be determined by using optical design knowledge and experience based on the requirements of different visual scenes for illumination, such as uniform illumination or spotlighting. The illumination simulation of the LED array according to the target spatial light distribution can be achieved by using optical simulation software to input LED parameters and target spatial light distribution for simulation, and adjusting the parameters multiple times to compare the simulation effect, selecting the result closest to the target distribution, and extracting the corresponding device parameters. The spatial light of the LED array configured using the preferred device parameters can be achieved by adjusting the arrangement of the LEDs according to the preferred device parameters, and changing the luminous intensity and direction of the LEDs by parameters such as the driving current. The target characteristics can be obtained by analyzing the size, movement speed, surface characteristics, etc. of the target in the scene using sensors and image recognition technology. The pulse width configuration strategy can be based on the influence of the target characteristics and the pulse width on the illumination effect, such as using short pulses for fast moving targets, to formulate a suitable pulse width adjustment scheme, including pulse width value, frequency, etc.

[0116] Furthermore, the embodiment of the present invention utilizes the initial control parameters to drive the preliminarily configured LED array to obtain a tunable composite spectrum about the RGB intelligent night vision light source, which can make the LED emit light according to the requirements of these parameters, thereby achieving preliminary adjustment of the spectral characteristics of the light source, thereby enhancing the contrast between the target and the background, thereby improving the detection and recognition capabilities of the target.

[0117] The tunable composite spectrum refers to a spectrum of composite light whose spectral composition and characteristics can be flexibly adjusted.

[0118] Optionally, the preliminarily configured LED array may be driven by a pulse generator in combination with the initial control parameters.

[0119] S3. Query the light reflection target in the light source construction area, illuminate the light reflection target with the tunable composite spectrum, collect light reflection data of the light reflection target, identify the data sparsity of the light reflection data, configure the observation matrix of the light reflection data based on the data sparsity, construct high dynamic range spectral cube data of the light reflection data with the observation matrix, construct an illumination-imaging joint optimization model for the light source construction area, optimize the high dynamic range spectral cube data with the illumination-imaging joint optimization model, and obtain target data.

[0120] The embodiment of the present invention can accurately know the situation of the light reflective target by querying the light reflective target in the light source construction area, and then adjust the irradiation direction, intensity and other parameters of the tunable composite spectrum in a targeted manner to ensure that the spectrum can effectively irradiate the target, avoid wasting light resources, and improve the efficiency and quality of subsequent data acquisition.

[0121] Optionally, the light reflective target can utilize optical sensors, radars and other detection equipment in combination with image recognition algorithms to scan and monitor the light source construction area to determine information such as the position, shape and distribution of target objects that reflect light.

[0122] The embodiment of the present invention uses the tunable composite spectrum to illuminate the light reflective target and collects light reflective data of the light reflective target, so as to facilitate the user to analyze information such as the spectral reflectance characteristics of the target.

[0123] Optionally, the use of the tunable composite spectrum to illuminate the light reflective target may be performed using a multi-spectral polarization imaging device.

[0124] Furthermore, the embodiments of the present invention can understand the inherent structure of the data by identifying the data sparsity of the light reflection data, which helps users to mine key information in the data.

[0125] As an embodiment of the present invention, the identifying data sparsity of the light reflection data includes: converting the light reflection data into vector data, arranging elements of vectors in the vector data to obtain arrangement elements, and calculating the Gini coefficient of the arrangement elements using the following formula:

[0126]

[0127] Among them, β represents the Gini coefficient, n represents the number of permutation elements, represents the absolute mean of the array elements, i represents the search variable of the array elements, x i Represents the i-th element in the array;

[0128] Based on the Gini coefficient, the data sparsity of the light reflection data is analyzed.

[0129] Optionally, the light reflection data is converted into vector data according to the storage form and structure of the light reflection data. If the light reflection data is multidimensional data (such as a matrix, a tensor, etc.), it can be converted into one-dimensional vector data in a specific manner (such as expanding by rows or columns). The arranged elements can be arranged in ascending or descending order according to the absolute value of the elements. The data sparsity of the light reflection data analyzed based on the Gini coefficient can be analyzed and judged according to the value of the calculated Gini coefficient and the standard of measuring data sparsity of the Gini coefficient (the closer the value is to 1, the sparser the data is).

[0130] The embodiment of the present invention can reduce the dimension of data and the computational complexity by configuring the observation matrix of the light reflection data based on the data sparsity, thereby improving the accuracy and reliability of data processing.

[0131] The measurement matrix refers to a mathematical tool used to convert a high-dimensional original signal into a low-dimensional observation value.

[0132] Optionally, the observation matrix can be obtained by analyzing the sparse structure and degree of the light reflection data to clarify its sparse distribution and the proportion of non-zero elements in each dimension, and then selecting the observation matrix type based on the analysis results, such as the sparse partial Fourier matrix of the data frequency domain, or a Gaussian random matrix.

[0133] Furthermore, the embodiment of the present invention can facilitate users to more comprehensively analyze the spectral characteristics and spatial distribution of light reflective targets by constructing high dynamic range spectral cube data of the light reflective data using the observation matrix.

[0134] The high dynamic range spectral cube data refers to a three-dimensional data structure with important applications in the fields of spectral analysis and imaging, which contains rich spatial, spectral and intensity information.

[0135] As an embodiment of the present invention, the use of the observation matrix to construct high dynamic range spectral cube data of the light reflection data includes: performing data preprocessing on the light reflection data to obtain preprocessed data, projecting the preprocessed data into the observation matrix to obtain projection data, identifying dimensional information of the projection data, constructing a spectral cube architecture of the projection data based on the dimensional information, filling the spectral cube architecture with data using the projection data to obtain a filled spectral cube, reconstructing the light reflection data to obtain reconstructed data, filling the reconstructed data into the filled spectral cube to obtain spectral cube data, and performing a posteriori processing on the spectral cube data to obtain high dynamic range spectral cube data.

[0136] Optionally, the pre-processed data can use a filtering algorithm to remove noise in the light reflection data and correct possible deviations in the data, such as by performing wavelength calibration on the spectral data. The projection data can be obtained by organizing the pre-processed data into a suitable matrix or vector form, performing matrix multiplication with the observation matrix, and obtaining it through this linear transformation. The dimensional information can be determined by analyzing the structure of the projection data to determine its quantity and range in spatial, spectral and other dimensions, such as determining the number of bands contained in the data, the number of spatial pixels, etc. The spectral cube architecture can define the spatial dimension, spectral dimension and possible other dimensions (such as time dimension, etc.) based on the identified dimensional information to build a three-dimensional data structure framework. The reconstructed data can be obtained by using compressed sensing or other suitable reconstruction algorithms, combined with the observation matrix and projection data, to recover the original light reflection data. The high dynamic range spectral cube data can be obtained by performing processing operations such as enhancing contrast and removing outliers on the spectral cube data to expand the dynamic range of the data.

[0137] The embodiment of the present invention can coordinate the mutual influence relationship between the lighting and imaging processes by constructing the lighting-imaging joint optimization model of the light source construction area, thereby giving full play to the performance advantages of the light source and the imaging system to improve the quality of target imaging, reduce noise and interference, and provide a more reliable foundation for subsequent data processing and analysis, while also improving the efficiency and performance of the entire system.

[0138] The lighting-imaging joint optimization model refers to a mathematical model that comprehensively considers the characteristics of the lighting system and the imaging system to achieve optimal overall performance.

[0139] As an embodiment of the present invention, the construction of the lighting-imaging joint optimization model of the light source construction area includes: querying the lighting-imaging requirements of the light source construction area, collecting system data of the light source construction area based on the lighting-imaging requirements, identifying the light source characteristics of the light source construction area, and constructing the lighting-imaging model architecture of the light source construction area based on the light source characteristics. Using the lighting-imaging requirements, constructing an optimization objective function of the lighting-imaging model architecture, analyzing the physical limitations of the lighting-imaging device in the light source construction area to construct the constraints of the lighting-imaging model architecture, deploying the lighting-imaging model architecture based on the optimization objective function and the constraints to obtain a lighting-imaging model, optimizing the lighting-imaging model using a genetic algorithm to obtain an optimized model, collecting real-time environmental data of the light source construction area to perform model verification on the optimization model to obtain verification parameters, performing index evaluation on the verification parameters, and obtaining a lighting-imaging joint optimization model when the evaluation result of the index evaluation meets the preset result.

[0140] Among them, the lighting-imaging requirements refer to the specific goals that the model wants to achieve, such as improving imaging clarity, enhancing target recognition accuracy, etc., and the system data refers to the environmental data of the lighting system, imaging system and light source construction area.

[0141] Optionally, the lighting-imaging requirements can be obtained by querying the user's requirements for the image clarity and enhanced target recognition accuracy of the light source construction area. The system data can be obtained by collecting the parameters of the light source in the light source construction area, including spectral distribution, luminous intensity and spatial distribution pattern; camera parameters such as resolution, sensitivity, noise level and optical characteristics of the lens (focal length, aperture, etc.); environmental data such as the size, shape and reflectivity distribution of the area. The light source characteristics can be obtained by analyzing the spectrum, luminous intensity and divergence angle of the light source through a spectrometer. The lighting-imaging model architecture can be constructed using a deep learning model based on light source characteristics and optical principles, and the optimization objective function can select appropriate evaluation indicators to measure imaging quality or information acquisition effects according to application requirements. Common indicators include image clarity (such as edge gradient, information entropy), contrast, color accuracy, target recognition accuracy, etc., and then the evaluation indicators are combined into an objective function, which can be linear or nonlinear. For example, for an application that focuses on both image clarity and contrast, the clarity index and the contrast index can be obtained by weighted summation. The constraints can take into account the physical limitations of the lighting system and the imaging system, such as the maximum power of the light source, the maximum frame rate of the camera, and the dynamic range of the sensor, etc., and the specific settings need to be combined with the actual application. The lighting-imaging model can be obtained by integrating the objective function and the constraints into the model architecture. The optimization model can be obtained by setting the parameters of the genetic algorithm, such as the population size, the number of iterations, etc., and simulating natural selection and genetic mechanisms to find the optimal solution of the model. The process of evaluating the index of the verification parameters is: first, according to the specific application requirements of the lighting-imaging system and the goals set when building the model, determine the specific indicators for evaluation. For example, in security monitoring applications, it may include image clarity (such as edge sharpness, resolution, etc.), contrast (brightness difference between the target and the background), color reproduction (the degree of consistency between the image color and the actual scene color), noise level (random interference signals in the image), etc.; in industrial detection, it may pay more attention to indicators such as the recognition accuracy of the target, the sensitivity and specificity of defect detection, and then set a specific quantitative calculation method for each indicator. For example, image clarity can be measured by calculating the gradient amplitude of the image, contrast can be calculated by dividing the difference in mean brightness between the target area and the background area by the sum of the mean brightness, color reproduction can be evaluated by comparing the color difference between the image and the standard color sample, etc. Finally, the verification parameters are compared with the preset lighting-imaging indicators.

[0142] The embodiment of the present invention optimizes the high dynamic range spectral cube data by using the illumination-imaging joint optimization model to obtain target data, which can improve the quality of analysis data and thus more accurately reflect the true characteristics of the light reflection target.

[0143] Optionally, the process of the target data is: inputting the high dynamic range spectral cube data into the lighting-imaging joint optimization model, adjusting the data in spatial and spectral dimensions according to the optimal parameter combination of lighting and imaging in the model, such as correcting the problem of uneven lighting, and then enhancing the data according to the optimization rules of the model to improve image clarity, contrast and recognition of spectral features and reduce noise and interference.

[0144] S4. Utilize the target data to identify the surface material characteristics and micro-motion features of the light reflective target, utilize the surface material characteristics and the micro-motion features to construct a scene semantic vector of the light source construction area, utilize the scene semantic vector in combination with a preconfigured strategy network to construct a light source control strategy for the preliminary configured LED array, construct a frequency domain-spatial domain joint optimization matrix for the light source control strategy, perform non-negative tensor decomposition on the frequency domain-spatial domain joint optimization matrix to obtain optimization unit parameters for the preliminary configured LED array, perform equipment deployment on the preliminary configured LED array based on the optimization unit parameters to obtain a target LED array, and utilize the target LED array to construct an RGB intelligent night vision light source for the light source construction area.

[0145] The embodiment of the present invention can understand the important attributes of the light reflective target by using the target data to identify the surface material characteristics and micro-motion characteristics of the light reflective target, thereby providing a basis for the subsequent construction of the scene semantic vector.

[0146] As an embodiment of the present invention, the method of using the target data to identify the surface material characteristics and micro-motion characteristics of the light reflective target includes: extracting spectral data about the light reflective target in the target data, analyzing the spectral characteristics of the light reflective target at different wavelengths using the spectral data, collecting image information of the light reflective target, constructing a grayscale co-occurrence matrix of the light reflective target, analyzing the texture information of the light reflective target using the grayscale co-occurrence matrix based on the image information, identifying the surface material characteristics of the light reflective target using the spectral characteristics and the texture information, collecting an inter-frame difference image of the light reflective target, calculating the optical flow field of the inter-frame difference image, and identifying the micro-motion characteristics of the light reflective target using the optical flow field.

[0147] Among them, the inter-frame difference image refers to an image obtained by subtracting two adjacent frame images in video or continuous image sequence processing, and the optical flow field refers to the movement speed and direction of pixel points in the image on the image plane over time.

[0148] Optionally, the spectral features can be identified by processing the extracted spectral data and analyzing its reflectivity, absorptivity, etc. at various wavelengths. The image information can be obtained by an imaging device (such as a camera, etc.). The grayscale co-occurrence matrix can be constructed by graying the collected image, and then according to the definition and calculation method of the grayscale co-occurrence matrix, the frequency of occurrence of different grayscale pixel pairs in the image is counted to construct the corresponding grayscale co-occurrence matrix. The texture information can be analyzed by calculating texture feature parameters such as contrast, correlation, energy, and entropy based on the grayscale co-occurrence matrix. The surface material characteristics can compare the obtained spectral features and texture information with the known feature database of different materials to comprehensively judge the surface material of the light-reflective target. The inter-frame difference image can be obtained by continuously acquiring multiple frames of images of the light-reflective target, and then subtracting two adjacent frames of images to obtain an inter-frame difference image. The optical flow field can be obtained by using an optical flow calculation algorithm to calculate the movement speed and direction of each pixel in the inter-frame difference image. The micro-motion feature can be obtained by analyzing the optical flow vector of the area where the light-reflective target is located in the optical flow field to determine the micro-motion direction, speed, and other characteristics of the light-reflective target.

[0149] Furthermore, the embodiment of the present invention can convert complex scene information into mathematical form by using the surface material characteristics and the micro-motion characteristics to construct the scene semantic vector of the light source construction area, which is convenient for computer processing and analysis. The scene semantic vector refers to a mathematical expression that represents specific scene information in the form of a numerical vector.

[0150] Optionally, the process of constructing the scene semantic vector of the light source construction area using the surface material characteristics and the micro-motion characteristics is as follows: first, the surface material characteristics and micro-motion characteristics of the light reflective target are quantified and encoded respectively, and an index table is established for the material to encode the material type and quantify the attributes, and the micro-motion characteristics are divided into direction interval encoding, normalized speed, extraction frequency and other quantizations; then, the quantized features of the two are combined into a feature vector; then, relevant information such as the size, shape, background environment, etc. of the light source construction area is considered and quantified and encoded, and merged with the feature vector of the light reflective target; finally, the merged vector is subjected to dimensionality reduction processing to remove redundancy and obtain the scene semantic vector.

[0151] The embodiment of the present invention utilizes the scene semantic vector in combination with a preconfigured strategy network to construct a light source control strategy for the preliminary configuration of the LED array, which can dynamically adjust the light source parameters of the LED array according to the characteristics of the actual scene, improve the utilization efficiency and lighting effect of the light source, and meet the needs of different scenes.

[0152] The preconfigured strategy network refers to a network model that has been structured and trained before performing a specific task (such as building a light source control strategy). It is used in the task to generate corresponding strategies based on input information and can be constructed using a decision tree model.

[0153] Optionally, the process of using the scene semantic vector in combination with a preconfigured strategy network to construct the light source control strategy for the preliminary configured LED array is as follows: preprocessing the scene semantic vector and inputting it into the strategy network, then using the strategy network to analyze and calculate the input data, inferring and outputting the light source control strategy for the preliminary configured LED array, which includes control parameters such as brightness, color, and flickering frequency, and then evaluating the output strategy to check whether it meets the restrictions and requirements, and optimizing and adjusting the strategy and strategy network parameters through simulation or testing when necessary.

[0154] Furthermore, the embodiment of the present invention can find the optimal combination of light sources in the frequency domain and the spatial domain by constructing the frequency domain-spatial domain joint optimization matrix of the light source control strategy, thereby further improving the performance and effect of the light source.

[0155] The frequency domain-spatial domain joint optimization matrix refers to a mathematical tool used to comprehensively optimize light source control strategies to achieve the best lighting effect in a specific scene.

[0156] As an embodiment of the present invention, constructing the frequency domain-spatial domain joint optimization matrix of the light source control strategy includes: querying the frequency domain range and spatial domain range of the light source construction area corresponding to the light source control strategy, constructing the frequency domain matrix of the light source construction area based on the frequency domain range, constructing the spatial domain matrix of the light source construction area based on the spatial domain range, constructing the frequency domain-spatial domain joint matrix of the light source control strategy using the frequency domain matrix and the spatial domain matrix, filling elements of the frequency domain-spatial domain joint matrix using matrix information of the frequency domain matrix and the spatial domain matrix to obtain a filled matrix, defining a matrix optimization function of the filled matrix, and using the matrix optimization function to perform matrix optimization on the filled matrix to obtain a frequency domain-spatial domain joint optimization matrix.

[0157] Among them, the matrix optimization function refers to a function with elements of the frequency domain-spatial domain joint matrix as independent variables, which maps the values ​​of the matrix elements to a scalar value, which represents a certain quantitative evaluation of the comprehensive performance of the light source control strategy in the frequency domain and spatial domain under the current matrix element values.

[0158] Optionally, the frequency domain range can be obtained by taking the required flicker frequency range as the frequency domain range according to the application scenario and requirements of the light source construction area, and the airspace range can be determined by measuring the length, width, height and grid division accuracy of the light source construction area. The frequency domain matrix can be divided into frequency intervals according to the frequency domain range, the matrix rows represent different frequency intervals, the columns correspond to different light sources or light source groups, and the element values ​​are preliminarily set according to the light source flicker frequency information. The airspace matrix can be divided into spatial grids according to the airspace range, the matrix elements correspond to the spatial positions, and the element values ​​are set according to the lighting requirements of each position, the contribution of the light source to the lighting of the position, etc. The frequency domain-airspace joint matrix can be obtained by creating a three-dimensional matrix with two dimensions of frequency domain and airspace. The filling matrix can be combined with the frequency domain and airspace matrix information, considering the effects of different frequency lights superimposed in space, interference, etc., and calculating and filling the joint matrix elements. The matrix optimization function can define the objective function by comprehensively considering the lighting uniformity, energy consumption and other factors of the intelligent light source, such as minimizing energy consumption under the premise of meeting the lighting intensity.

[0159] Furthermore, the embodiment of the present invention can simplify and decompose the complex joint optimization matrix by performing non-negative tensor decomposition on the frequency-domain-spatial-domain joint optimization matrix to obtain the optimization unit parameters for the preliminary configuration of the LED array, and extract more specific and more operational optimization unit parameters, thereby providing precise guidance for subsequent equipment deployment.

[0160] Optionally, the process of performing non-negative tensor decomposition on the frequency-domain-spatial domain joint optimization matrix to obtain optimized unit parameters for the preliminary configuration of the LED array is as follows: first, set parameters such as the rank of the non-negative tensor decomposition (i.e., the order of the low-dimensional tensor after decomposition), and select a suitable decomposition algorithm, such as the alternating least squares method; then, regard the frequency-domain-spatial domain joint optimization matrix as a tensor, iteratively calculate it using the selected algorithm, and decompose it into multiple non-negative low-dimensional tensors; finally, extract parameters related to the preliminary configuration of the LED array from the decomposed low-dimensional tensor, such as the brightness and frequency of each LED, as optimized unit parameters.

[0161] The embodiment of the present invention performs equipment deployment on the preliminary configured LED array based on the optimization unit parameters to obtain a target LED array, and can apply the theoretical optimization parameters to the actual LED array, thereby ensuring that the LED array can work according to the optimized strategy, achieve the best performance and effect of the light source, and improve the quality and efficiency of the entire lighting system.

[0162] Optionally, the process of performing equipment deployment on the preliminarily configured LED array based on the optimization unit parameters to obtain the target LED array is as follows: based on the optimization unit parameters, accurately adjust the operating parameters (such as brightness, flashing frequency, color, etc.) of each LED in the preliminarily configured LED array to make the state of each LED consistent with the parameter requirements; if the optimization unit parameters involve LED position information, the physical position of the LED in the array is rearranged to ensure that its spatial distribution complies with the optimization plan; after completing the adjustment and layout, the deployed LED array is tested to check whether various indicators meet the standards; if there are deviations, calibration is performed to finally obtain the target LED array.

[0163] Furthermore, the embodiment of the present invention can construct an intelligent light source that can meet night vision needs by utilizing the target LED array to construct the RGB intelligent night vision light source in the light source construction area, and can provide customized lighting solutions according to different scenes and target characteristics, thereby improving the effect of night observation and monitoring.

[0164] Embodiment 2:

[0165] like Figure 2 The figure shows a functional module diagram of a system for constructing an RGB intelligent night vision light source according to the present invention.

[0166] The construction system 200 of an RGB intelligent night vision light source of the present invention can be installed in an electronic device. According to the functions to be implemented, the construction system of an RGB intelligent night vision light source can include an initial parameter construction module 201, a device parameter configuration module 202, an environmental data optimization module 203 and an intelligent night vision light source construction module 204. The module of the present invention can also be referred to as a unit, which refers to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, which are stored in the memory of the electronic device.

[0167] In the embodiment of the present invention, the functions of each module / unit are as follows:

[0168] The initial parameter construction module 201 is used to collect the ambient light intensity, spectral distribution and dynamic target thermal radiation data of the light source construction area corresponding to the RGB intelligent night vision light source, obtain environmental perception data, construct the environmental feature tensor of the environmental perception data, and construct the initial control parameters of the RGB intelligent night vision light source using a double delay network based on the environmental feature tensor;

[0169] The device parameter configuration module 202 is used to query the LED array of the RGB intelligent night vision light source, configure the spatial light and pulse width of the LED array, obtain a preliminary configuration LED array, drive the preliminary configuration LED array using the initial control parameters, and obtain a tunable composite spectrum for the RGB intelligent night vision light source;

[0170] The environmental data optimization module 203 is used to query the light reflection target of the light source construction area, illuminate the light reflection target with the tunable composite spectrum, collect light reflection data of the light reflection target, identify the data sparsity of the light reflection data, configure the observation matrix of the light reflection data based on the data sparsity, construct high dynamic range spectral cube data of the light reflection data with the observation matrix, construct an illumination-imaging joint optimization model of the light source construction area, and optimize the high dynamic range spectral cube data with the illumination-imaging joint optimization model to obtain target data;

[0171] The intelligent night vision light source construction module 204 is used to use the target data to identify the surface material characteristics and micro-motion features of the light reflective target, use the surface material characteristics and the micro-motion features to construct a scene semantic vector of the light source construction area, use the scene semantic vector in combination with a preconfigured strategy network to construct a light source control strategy for the preliminary configured LED array, construct a frequency domain-spatial domain joint optimization matrix for the light source control strategy, perform non-negative tensor decomposition on the frequency domain-spatial domain joint optimization matrix to obtain optimization unit parameters for the preliminary configured LED array, perform equipment deployment on the preliminary configured LED array based on the optimization unit parameters to obtain a target LED array, and use the target LED array to construct an RGB intelligent night vision light source for the light source construction area.

[0172] In detail, each module in the construction system 200 of an RGB intelligent night vision light source in the embodiment of the present invention is used in the same manner as described above. Figure 1 The same technical means as the method for constructing an RGB intelligent night vision light source described in the text can produce the same technical effects, so I will not go into details here.

[0173] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0174] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for constructing an RGB intelligent night vision light source, characterized in that: The method comprises: Collecting ambient light intensity, spectral distribution and dynamic target thermal radiation data of the light source construction area corresponding to the RGB intelligent night vision light source to obtain environmental perception data, constructing an environmental feature tensor of the environmental perception data, and based on the environmental feature tensor, using a double delay network to construct initial control parameters of the RGB intelligent night vision light source; Querying the LED array of the RGB intelligent night vision light source, configuring the spatial light and pulse width of the LED array, obtaining a preliminary configured LED array, and driving the preliminary configured LED array using the initial control parameters to obtain a tunable composite spectrum of the RGB intelligent night vision light source; Query the light reflection target of the light source construction area, illuminate the light reflection target with the tunable composite spectrum, collect light reflection data of the light reflection target, identify the data sparsity of the light reflection data, configure the observation matrix of the light reflection data based on the data sparsity, construct high dynamic range spectral cube data of the light reflection data with the observation matrix, construct an illumination-imaging joint optimization model of the light source construction area, optimize the high dynamic range spectral cube data with the illumination-imaging joint optimization model, and obtain target data; The target data is used to identify the surface material characteristics and micro-motion features of the light reflective target, and the surface material characteristics and the micro-motion features are used to construct a scene semantic vector of the light source construction area. The scene semantic vector is combined with a preconfigured strategy network to construct a light source control strategy for the preliminary configured LED array, and a frequency domain-spatial domain joint optimization matrix of the light source control strategy is constructed. The frequency domain-spatial domain joint optimization matrix is ​​non-negative tensor decomposition is performed on the frequency domain-spatial domain joint optimization matrix to obtain optimization unit parameters for the preliminary configured LED array, and equipment deployment is performed on the preliminary configured LED array based on the optimization unit parameters to obtain a target LED array, and the target LED array is used to construct an RGB intelligent night vision light source for the light source construction area.

2. The method for constructing an RGB intelligent night vision light source as claimed in claim 1, characterized in that: The method of constructing the initial control parameters of the RGB intelligent night vision light source based on the environmental feature tensor using a double delay network includes: Deploy a dual-delay network of the RGB intelligent night vision light source; Using the environmental feature tensor to construct a policy network and a value network of the dual-delay network to obtain a deployed dual-delay network; Initializing network parameters of the deployed double-delay network, and constructing an experience playback buffer of the deployed double-delay network to obtain a target double-delay network; After inputting the environment feature tensor into the policy network of the target dual-delay network, sampling buffer data from the experience replay buffer to obtain sampled data; The sampled data is used to update the value network in the target double-delay network to obtain an updated double-delay network. network; After the environmental feature tensor is input into the updated double-delay network, the output data of the updated double-delay network is collected to obtain initial control parameters.

3. A method for constructing an RGB intelligent night vision light source as claimed in claim 2, characterized in that: The method of updating the value network in the target double-delay network by using the sampled data to obtain an updated double-delay network includes: Extracting sampling actions and sampling states from the sampling data; After the sampling action and the sampling state are input into the value network, the target value of the value network is calculated using the following formula: y i =r i +γmin j＝1,2 Q′(s i+1 ,π′(s i+1 );θ′ j ) Among them, y i represents the target value, r i represents the i-th reward value in the sampled data, γ represents the discount factor of the value network, Q′ represents the target value network corresponding to the value network, and s i+1 represents the i+1th sampling state, θ′ j represents the jth parameter of the target value network; Based on the target value, the mean square error of the value network is calculated using the following formula: Among them, L(θ) represents the mean square error, N represents the number of sampled data, and y i represents the target value, s i represents the i-th sampling state, a i represents the i-th sampled action, θ represents the parameters of the value network; Based on the mean square error, the value network in the target double-delay network is updated by using a gradient descent method to obtain an updated double-delay network.

4. The method for constructing an RGB intelligent night vision light source as claimed in claim 1, characterized in that: The configuring the spatial light and pulse width of the LED array to obtain a preliminary configuration of the LED array includes: identifying a visual scene of the LED array; Based on the visual scene, determining a target spatial light distribution of the LED array; Performing illumination simulation on the LED array according to the target spatial light distribution, and extracting preferred device parameters corresponding to preferred simulation results in the illumination simulation; configuring the spatial light of the LED array using the preferred device parameters; querying a visual scene of the LED array and identifying target characteristics in the visual scene; Based on the target characteristics, construct a pulse width configuration strategy for the LED array; After the pulse width of the LED array is configured using the pulse width configuration strategy, a preliminary configured LED array is obtained.

5. The method for constructing an RGB intelligent night vision light source as claimed in claim 1, characterized in that: The identifying data sparsity of the light reflection data includes: Converting the light reflection data into vector data; Arrange the elements of the vectors in the vector data to obtain arrangement elements; The Gini coefficient of the arrangement element is calculated using the following formula: Among them, β represents the Gini coefficient, n represents the number of permutation elements, represents the absolute mean of the array elements, i represents the search variable of the array elements, x i Represents the i-th element in the array; Based on the Gini coefficient, the data sparsity of the light reflection data is analyzed.

6. The method for constructing an RGB intelligent night vision light source as claimed in claim 1, characterized in that: The method of constructing high dynamic range spectral cube data of the light reflection data using the observation matrix includes: Performing data preprocessing on the light reflection data to obtain preprocessed data; Projecting the preprocessed data into the measurement matrix to obtain projection data; Identifying dimensional information of the projection data; constructing a spectral cube architecture of the projection data based on the dimensional information; Using the projection data to fill the spectral cube architecture with data to obtain a filled spectral cube; Reconstructing the light reflection data to obtain reconstructed data; Filling the reconstructed data into the filled spectral cube to obtain spectral cube data; The spectral cube data is processed a posteriori to obtain high dynamic range spectral cube data.

7. The method for constructing an RGB intelligent night vision light source as claimed in claim 1, characterized in that: The constructing of the illumination-imaging joint optimization model of the light source construction area includes: Query the lighting-imaging requirements of the light source construction area; Collecting system data of the light source construction area based on the lighting-imaging requirements; Identifying light source characteristics of the light source construction area; Based on the light source characteristics, construct an illumination-imaging model framework of the light source construction area; Using the lighting-imaging requirements, constructing an optimization objective function of the lighting-imaging model architecture; Analyzing the physical limitations of the illumination-imaging device in the light source construction area to construct the constraints of the illumination-imaging model architecture; Based on the optimization objective function and the constraint conditions, the illumination-imaging model architecture is deployed to obtain an illumination-imaging model; Optimizing the illumination-imaging model using a genetic algorithm to obtain an optimized model; Collecting real-time environmental data of the light source construction area to perform model verification on the optimization model and obtain verification parameters; An index evaluation is performed on the verification parameters, and when an evaluation result of the index evaluation meets a preset result, an illumination-imaging joint optimization model is obtained.

8. The method for constructing an RGB intelligent night vision light source as claimed in claim 1, characterized in that: The method of using the target data to identify the surface material characteristics and micro-motion characteristics of the light reflective target includes: Extracting spectral data about the light reflective target from the target data; Analyzing the spectral characteristics of the light reflective target at different wavelengths using the spectral data; Collecting image information of the light reflection target; Constructing a gray-level co-occurrence matrix of the light-reflecting target; Analyzing texture information of the light reflective target using the gray level co-occurrence matrix based on the image information; Identifying the surface material characteristics of the light reflective target using the spectral characteristics and the texture information; Collecting an inter-frame difference image of the light reflective target; Calculating the optical flow field of the inter-frame difference image; The optical flow field is used to identify micro-motion features of the optical reflective target.

9. The method for constructing an RGB intelligent night vision light source according to claim 1, characterized in that: The constructing of the frequency domain-spatial domain joint optimization matrix of the light source control strategy includes: Query the frequency domain range and spatial domain range of the light source construction area corresponding to the light source control strategy; Constructing a frequency domain matrix of the light source construction area based on the frequency domain range; Constructing a spatial matrix of the light source construction area based on the spatial range; Using the frequency domain matrix and the spatial domain matrix, construct a frequency domain-spatial domain joint matrix of the light source control strategy; Filling the frequency-domain-spatial-domain joint matrix with elements using matrix information of the frequency-domain matrix and the spatial-domain matrix to obtain a filled matrix; defining a matrix optimization function for the filling matrix; The matrix optimization function is used to perform matrix optimization on the filling matrix to obtain a frequency domain-spatial domain joint optimization matrix.

10. A system for constructing an RGB intelligent night vision light source, characterized in that: The system comprises: An initial parameter construction module is used to collect ambient light intensity, spectral distribution and dynamic target thermal radiation data of the light source construction area corresponding to the RGB intelligent night vision light source, obtain environmental perception data, construct an environmental feature tensor of the environmental perception data, and construct the initial control parameters of the RGB intelligent night vision light source using a double delay network based on the environmental feature tensor; A device parameter configuration module, used for querying the LED array of the RGB intelligent night vision light source, configuring the spatial light and pulse width of the LED array, obtaining a preliminary configured LED array, and driving the preliminary configured LED array using the initial control parameters to obtain a tunable composite spectrum of the RGB intelligent night vision light source; An environmental data optimization module is used to query the light reflection target in the light source construction area, illuminate the light reflection target with the tunable composite spectrum, collect light reflection data of the light reflection target, identify the data sparsity of the light reflection data, configure the observation matrix of the light reflection data based on the data sparsity, construct high dynamic range spectral cube data of the light reflection data with the observation matrix, construct an illumination-imaging joint optimization model for the light source construction area, and optimize the high dynamic range spectral cube data with the illumination-imaging joint optimization model to obtain target data; The intelligent night vision light source construction module is used to use the target data to identify the surface material characteristics and micro-motion features of the light reflective target, use the surface material characteristics and the micro-motion features to construct a scene semantic vector of the light source construction area, use the scene semantic vector in combination with a preconfigured strategy network to construct a light source control strategy for the preliminary configured LED array, construct a frequency domain-spatial domain joint optimization matrix for the light source control strategy, perform non-negative tensor decomposition on the frequency domain-spatial domain joint optimization matrix to obtain optimization unit parameters for the preliminary configured LED array, perform equipment deployment on the preliminary configured LED array based on the optimization unit parameters to obtain a target LED array, and use the target LED array to construct an RGB intelligent night vision light source for the light source construction area.

Images