Self-adaptive control method and device for supplementary light source of industrial camera

Abstract

The invention relates to an adaptive control method and device for a supplementary light source of an industrial camera, and the method comprises the steps: obtaining scene perception data of the industrial camera, carrying out the real-time classification and environment parameter analysis of a shooting scene based on the scene perception data, and obtaining scene feature information; performing image quality evaluation on the scene feature information, and comparing with a preset image quality standard to obtain an image quality matching degree; performing control parameter analysis based on the image quality matching degree and the scene feature information to obtain an initial light supplementing control parameter; acquiring image data of the industrial camera, and performing brightness distribution and detail feature analysis on the image data to obtain an image analysis result; performing dynamic light source simulation on the scene feature information based on an image analysis result to obtain a corresponding dynamic supplementary light source parameter; and performing comprehensive processing on the initial light supplement control parameter, the image analysis result and the dynamic light supplement source parameter to obtain a light supplement source adjustment strategy. The problem that a traditional fixed light supplementing method is difficult to adapt to a complex environment can be effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial cameras, and in particular to a method and device for adaptively controlling a supplementary light source of an industrial camera. Background Art

[0002] Industrial cameras play an increasingly important role in modern manufacturing and quality control. With the rapid development of Industry 4.0 and smart manufacturing, the requirements for industrial camera image quality are also increasing. However, in practical applications, the shooting environment of industrial cameras is often complex and the lighting conditions are diverse, which poses a huge challenge to the clarity and accuracy of the image. Traditional industrial camera fill light methods usually use fixed fill light parameters, which are difficult to adapt to the needs of different scenes and environmental conditions. This single fill light strategy may lead to unstable image quality and affect subsequent image processing and analysis results. Therefore, how to realize the intelligent and adaptive control of industrial camera fill light sources to adapt to different shooting scenes and environmental conditions and improve the stability and reliability of image quality has become an important direction of current research. Summary of the invention

[0003] The main purpose of the present invention is to provide an industrial camera fill light source adaptive control method and device, which can effectively solve the problem that the traditional fixed fill light method is difficult to adapt to complex environments.

[0004] To achieve the above object, the present invention provides an industrial camera supplementary light source adaptive control method, comprising: Acquire scene perception data of the industrial camera, perform real-time classification of the shooting scene and environmental parameter analysis based on the scene perception data, and obtain scene feature information; Performing image quality evaluation on the scene feature information and comparing it with a preset image quality standard to obtain an image quality matching degree; Performing control parameter analysis based on the image quality matching degree and the scene feature information to obtain initial fill light control parameters; Acquire image data of the industrial camera, perform brightness distribution and detail feature analysis on the image data, and obtain corresponding image analysis results; Performing dynamic light source simulation on the scene feature information based on the image analysis result to obtain corresponding dynamic supplementary light source parameters; The initial fill light control parameter, the image analysis result and the dynamic fill light source parameter are comprehensively processed to obtain a fill light source adjustment strategy.

[0005] Furthermore, the scene perception data of the industrial camera is obtained, and the real-time classification of the shooting scene and environmental parameter analysis are performed based on the scene perception data to obtain scene feature information, including: Acquire the spectral data and depth data collected by the industrial camera, and integrate them to obtain the scene perception data; Performing high-order singular value decomposition on the scene perception data and extracting cross-modal correlation features to obtain a modal coupling matrix; Based on the modal coupling matrix, a scene is constructed to obtain a scene semantic graph; Performing Betti number sequence calculation on the scene semantic graph according to a preset continuous homology algorithm to obtain a scene topology feature vector; Performing multi-scale fractal dimension analysis based on the scene topology feature vector to obtain a scene complexity spectrum; Performing dynamic stability characteristic analysis based on the scene complexity spectrum to obtain scene dynamic characteristics; Symbolize the dynamic features of the scene to generate a symbol sequence of the scene; Calculating conditional entropy and mutual information based on the symbol sequence to obtain a scene uncertainty index; The scene topology feature vector, the scene dynamic feature and the scene uncertainty index are subjected to nonlinear mapping fusion to obtain the scene feature information.

[0006] Furthermore, the performing image quality assessment on the scene feature information and comparing it with a preset image quality standard to obtain an image quality matching degree includes: Calculating the brightness entropy value of the scene feature information to obtain a scene brightness entropy value; Performing brightness uniformity evaluation on the scene feature information according to the scene brightness entropy value to obtain a brightness uniformity score; Performing texture spectrum analysis on the scene feature information to obtain texture spectrum features; Performing detail retention evaluation on the scene feature information according to the texture spectrum feature to obtain a detail retention score; Performing color saturation calculation on the scene feature information to obtain a color saturation value; Performing a color restoration evaluation on the scene feature information according to the color saturation value to obtain a color restoration score; Comprehensively calculating the brightness uniformity score, the detail retention score, and the color reproduction score to obtain a comprehensive image quality score; A matching degree analysis is performed between the comprehensive image quality score and the image quality standard to obtain an image quality matching degree.

[0007] Furthermore, the performing control parameter analysis based on the image quality matching degree and the scene feature information to obtain initial fill light control parameters includes: Performing a nonlinear transformation of a hyperbolic tangent function on the image quality matching degree to obtain a quality matching curve; Extracting high-order statistical features and frequency domain features from the scene feature information, and constructing a multi-dimensional feature space to obtain a scene feature vector; Performing parameter analysis on the quality matching curve and the scene feature vector to obtain an initial fill light parameter set; Iteratively calculating the initial fill light parameter set according to a preset image clarity evaluation function to obtain optimized gradient information; The initial fill light parameter set is dynamically adjusted according to the optimization gradient information to obtain the initial fill light control parameters.

[0008] Furthermore, the acquiring of the image data of the industrial camera, performing brightness distribution and detail feature analysis on the image data, and obtaining corresponding image analysis results, includes: Performing multi-scale pyramid decomposition on the image data to obtain multi-scale frequency coefficients; Perform energy distribution analysis of each sub-band based on the multi-scale frequency coefficients to obtain a sub-band frequency domain feature map; Performing region value segmentation on the sub-band frequency domain feature map to obtain a significant region matrix; Divide the image data into regions according to the salient region matrix to obtain a target region; Performing local directional feature analysis on the target area to obtain local directional features; Calculating the direction consistency of regional pixels based on the local direction features to obtain a detail descriptor; Performing cluster analysis on the detail descriptors to obtain an image texture complexity index; Performing matrix construction based on the sub-band frequency domain feature map and the texture complexity index to obtain a constructed joint distribution matrix; Performing vector eigendecomposition on the joint distribution matrix to obtain an image matrix vector; An image quality score is performed based on the image matrix vector and the joint distribution matrix to obtain the image analysis result.

[0009] Furthermore, the performing dynamic light source simulation on the scene feature information based on the image analysis result to obtain corresponding dynamic supplementary light source parameters includes: Performing illumination non-uniformity evaluation on the image analysis result to obtain illumination uniformity information; Virtually adjusting the light source position of the scene feature information according to the illumination uniformity information to obtain multiple groups of virtual light source position parameters; Performing ray tracing calculation on each group of virtual light source position parameters to obtain corresponding illumination distribution data; Analyzing the object surface reflection characteristics of the scene feature information according to the illumination distribution data to obtain a surface reflectivity distribution map; Correcting the illumination distribution data based on the surface reflectivity distribution map to obtain corrected illumination distribution data; Performing light intensity optimization calculation on the corrected light distribution data to obtain optimized light intensity parameters; Generate multiple groups of candidate fill light solutions according to the optimized light intensity parameters and the virtual light source position parameters; Performing image quality prediction and evaluation on each group of candidate fill-light solutions to obtain corresponding image quality prediction indicators; Sorting and screening the candidate fill light solutions according to the image quality prediction index to obtain a screened fill light solution; The light source position and light intensity parameters are dynamically converted for the screened fill light scheme to obtain the dynamic fill light source parameters.

[0010] Furthermore, the comprehensive processing of the initial fill light control parameter, the image analysis result and the dynamic fill light source parameter to obtain a fill light source adjustment strategy includes: Performing fractal dimension analysis on the initial fill light control parameters to obtain a fill light complexity index; Constructing an optical flow field according to the image analysis results to obtain a scene dynamic optical flow field map; Performing tensor decomposition on the dynamic supplementary light source parameters to obtain multimodal light source features; A topological structure is constructed based on the fill light complexity index and the scene dynamic optical flow field map to obtain an adaptive fill light network structure; Mapping the multimodal light source characteristics onto the adaptive fill light network structure to obtain a preliminary fill light strategy; The fill light source control pulse sequence is optimized based on the preliminary fill light strategy to obtain the fill light source adjustment strategy.

[0011] Furthermore, mapping the multimodal light source characteristics onto the adaptive fill light network structure to obtain a preliminary fill light strategy includes: Performing feature vector decomposition on the multimodal light source features to obtain main components of light source features; Assigning node weights to the adaptive fill-light network structure according to the main components of the light source characteristics to obtain a weighted fill-light network; Performing graph analysis on the weighted light supplement network and extracting network topology features to obtain a network structure feature vector; Performing a tensor product operation on the network structure feature vector and the light source feature principal component to obtain a fused feature tensor; Performing nonlinear transformation on the fused feature tensor to obtain a candidate set of fill light strategies; The fill light strategy candidate set is screened according to a preset fill light sorting standard to obtain a corresponding screened fill light strategy set; The screened fill-light strategy set is integrated to obtain the preliminary fill-light strategy.

[0012] The present invention also provides an industrial camera supplementary light source adaptive control device, which is applied to any one of the above industrial camera supplementary light source adaptive control methods, comprising: An acquisition module, which is used to obtain scene perception data of an industrial camera, and to perform real-time classification of shooting scenes and environmental parameter analysis based on the scene perception data to obtain scene feature information; An analysis module, the analysis module is used to evaluate the image quality of the scene feature information and compare it with a preset image quality standard to obtain an image quality matching degree; An association module, the association module is used to perform control parameter analysis based on the image quality matching degree and the scene feature information to obtain initial fill light control parameters; A processing module, the processing module is used to obtain image data of the industrial camera, perform brightness distribution and detail feature analysis on the image data, and obtain corresponding image analysis results; A control module, the control module is used to perform dynamic light source simulation on the scene feature information based on the image analysis result to obtain corresponding dynamic supplementary light source parameters; An execution module is used to comprehensively process the initial fill light control parameters, the image analysis results and the dynamic fill light source parameters to obtain a fill light source adjustment strategy.

[0013] The present invention provides an industrial camera supplementary light source adaptive control method and device, which has the following beneficial effects: By acquiring scene perception data and performing real-time classification and environmental parameter analysis, the fill light requirements of different shooting scenes can be evaluated more accurately, thereby improving the accuracy of fill light control and providing a more reliable basis for image acquisition. By evaluating the image quality of scene feature information and comparing it with the preset standard, the refined management of fill light requirements for different scenes is achieved, which helps to achieve on-demand fill light and avoid waste of light source resources. Control parameter analysis based on image quality matching and scene feature information can ensure that the fill light system can operate efficiently under different environmental conditions, reduce unnecessary energy consumption, and improve the overall operation efficiency of the system. By analyzing the brightness distribution and detail characteristics of image data and performing dynamic light source simulation, a more reasonable fill light strategy can be formulated, thereby effectively utilizing light source resources and improving the stability of image quality. By comprehensively processing the initial fill light control parameters, image analysis results, and dynamic fill light source parameters, the fill light strategy can be flexibly adjusted according to the characteristics of different scenes and changes in demand, making the system more adaptable to diverse industrial application scenarios and effectively solving the problem that traditional fixed fill light methods are difficult to adapt to complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is a flow chart of an adaptive control method for supplementary light sources for industrial cameras provided by the present invention; Figure 2 The present invention provides a structural diagram of an adaptive control device for supplementary light sources for industrial cameras.

[0015] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0017] The present invention is further described below in conjunction with the accompanying drawings and specific implementation methods.

[0018] Reference Figure 1 As shown, the present invention provides an industrial camera supplementary light source adaptive control method, comprising: Step S1: Acquire scene perception data of the industrial camera, perform real-time classification of the shooting scene and environmental parameter analysis based on the scene perception data, and obtain scene feature information; Step S2: performing image quality evaluation on the scene feature information and comparing it with a preset image quality standard to obtain an image quality matching degree; Step S3: analyzing control parameters based on image quality matching and scene feature information to obtain initial fill light control parameters; Step S4: Acquire image data from an industrial camera, perform brightness distribution and detail feature analysis on the image data, and obtain corresponding image analysis results; Step S5: Perform dynamic light source simulation on the scene feature information based on the image analysis result to obtain corresponding dynamic fill light source parameters; Step S6: Comprehensively process the initial fill light control parameters, the image analysis results and the dynamic fill light source parameters to obtain a fill light source adjustment strategy.

[0019] Based on the above steps, the detailed process is as follows: Step S1: The industrial camera uses auxiliary equipment such as light sensors, temperature and humidity sensors, etc. to achieve comprehensive perception of the shooting environment. The light sensor can accurately measure key parameters such as ambient light intensity and color temperature, while the temperature and humidity sensor monitors the physical state of the environment. At the same time, the camera also needs to record spatiotemporal information such as shooting time and geographic location to provide basic data for subsequent scene analysis.

[0020] Real-time scene classification is to quickly classify the current shooting scene into different types, such as indoor / outdoor, bright / dark, static / dynamic, etc. This classification not only relies on the visual features of the image itself, but also comprehensively considers multi-dimensional information such as lighting and background. By establishing a scene feature library, the classification algorithm can be continuously optimized to improve recognition accuracy and real-time performance.

[0021] Environmental parameter analysis is an in-depth exploration and interpretation of scene perception data. The stability, change trend, color temperature and color rendering of the ambient light will be comprehensively evaluated, and the spatial distribution characteristics of the light will be calculated. Physical parameters such as temperature and humidity are also included in the analysis because they will significantly affect the image quality. Through multi-dimensional and multi-level comprehensive analysis, a more comprehensive and accurate description of the environment can be obtained.

[0022] The final output of scene feature information is a standardized feature vector, which integrates scene classification results, environmental parameter analysis, etc. This vector not only records the objective characteristics of the current scene, but also contains a deep understanding of the imaging environment.

[0023] Step S2: The brightness uniformity, contrast, clarity and color reproduction of image quality assessment are the core evaluation dimensions. Brightness uniformity reflects the consistency of brightness in each area of ​​the image, contrast reflects the layering and detail richness of the image, clarity measures the sharpness of the image, and color reproduction reflects the authenticity and saturation of the color.

[0024] The formulation of preset image quality standards needs to fully consider the needs of different application scenarios. For example, industrial inspection, medical imaging, security monitoring and other fields have different requirements for image quality. Therefore, the quality standards should be multi-level and adjustable. By setting the weight coefficient and threshold range of each indicator, a flexible and accurate evaluation system can be constructed.

[0025] The implementation process of image quality assessment is based on intelligent prediction of scene feature information. The possible image quality level will be estimated in advance based on scene type, environmental parameters, etc. By calculating the scores of various indicators and performing weighted synthesis, a detailed quality assessment result is generated.

[0026] The calculation of quality matching is to accurately compare the evaluation results with the preset standards. The matching degree of each indicator will be calculated separately, and the overall matching degree will be finally obtained. The matching degree is not only a numerical value, but also a quantitative description of the degree of deviation of the image quality from the ideal state. Low matching means that the imaging effect needs to be improved by means of fill light, etc.

[0027] Step S3: Fill light demand analysis is the premise of generating control parameters. The priority and urgency of fill light will be determined based on the quality matching degree, and the specific requirements of the scene characteristics for fill light will be comprehensively evaluated. The difficulty and feasibility of fill light vary significantly in different scenes.

[0028] The initial parameter calculation is an accurate derivation based on the previous analysis. The basic fill light requirements are calculated based on the scene characteristics, and then the fill light intensity is further adjusted based on the quality matching degree. The optimization of the fill light timing is also an important link, which needs to consider the response characteristics of the light source and the law of scene changes. The final parameter set should be both in line with theoretical expectations and practically operable.

[0029] Parameter feasibility verification is a key step to ensure the reliability of the fill light control solution. The generated parameters will be fully checked to see if they are within the range allowed by the device, the stability of the parameters will be evaluated, and the fill light effect will be predicted. Only parameters that have passed strict verification can be actually implemented, which is also an important guarantee for the reliability of the fill light system.

[0030] Step S4: The industrial camera will capture high-precision, high-resolution raw image data and ensure the integrity and authenticity of the data. The data acquisition process needs to consider key technical indicators such as the dynamic range and signal-to-noise ratio of the image sensor to ensure the quality of the input data. Establish a standardized data preprocessing process to perform necessary denoising and correction on the raw image.

[0031] Brightness distribution analysis is the core link of image quality assessment. Image processing algorithms will be used to analyze the brightness of the image in detail in multiple dimensions and scales. Through histogram analysis, regional brightness mapping and other technologies, the brightness spatial distribution characteristics of the image can be accurately depicted. Focus on the uniformity of brightness, gradient changes, and brightness contrast in key areas, providing an important basis for subsequent fill light decisions.

[0032] Detail feature analysis is a fine mining of deep information of images. It extracts key features such as texture, edge, and texture of the image. The analysis will cover multiple scales and abstract levels, focusing not only on macroscopic structures but also digging deep into microscopic details. Through feature mapping and feature importance evaluation, it fully understands the structural complexity and information richness of the image.

[0033] The output of the image analysis result is a multi-dimensional feature vector. It comprehensively reflects the key information of the image in terms of brightness distribution, detail features, texture structure, etc. This vector is not only a quantitative description of the image, but also a bridge between image perception and fill light control. It will provide accurate input for subsequent dynamic light source simulation and fill light strategy formulation.

[0034] Step S5: Through physical optical simulation and computer rendering technology, the lighting effects under different lighting solutions can be accurately predicted. The simulation process takes into account multi-dimensional factors such as light source type, lighting angle, reflection characteristics, etc., and strives to obtain the lighting reconstruction that is closest to the real environment.

[0035] The dynamic association of scene feature information is the key to the simulation process. The image analysis results are deeply integrated with the scene feature information obtained in the early stage to establish a dynamic and real-time scene optical model. Through machine learning algorithms, the accuracy and adaptability of light source simulation can be continuously optimized. The model will be able to quickly respond to scene changes and realize intelligent prediction and simulation of the lighting environment.

[0036] The generation of dynamic fill light source parameters is a multi-objective optimization problem. Multiple constraints such as image quality improvement, energy consumption control, and light source life are considered comprehensively, and the optimal fill light parameters are generated through intelligent algorithms. The parameters include fill light intensity, color temperature, spatial distribution, and other dimensions. The optimization goal is to achieve efficient and energy-saving operation of fill light while ensuring image quality.

[0037] Verification of parameter accuracy is a key step to ensure the reliability of simulation results. The scientificity and feasibility of dynamic supplementary light source parameters will be comprehensively evaluated through model comparison, experimental verification and other methods. This process is not only a technical test of the parameters, but also a continuous optimization and improvement of the entire light source simulation method.

[0038] Step S6: Comprehensively process the initial fill light control parameters, the image analysis results and the dynamic fill light source parameters to obtain a fill light source adjustment strategy.

[0039] Comprehensive processing is the integration and optimization stage of the entire fill light control solution. The initial control parameters, image analysis results and dynamic fill light source parameters obtained in the early stage are integrated in multiple dimensions and levels. The comprehensive processing results are obtained by prioritizing and co-optimizing various parameters.

[0040] The formulation of the supplementary light source adjustment strategy is a dynamic and adaptive process. Based on the comprehensive processing results, a multi-level and evolvable adjustment framework is constructed. The strategy includes not only static parameter settings, but also a dynamic adjustment mechanism based on real-time feedback. Through closed-loop control and self-learning algorithms, intelligent adaptation to complex and changing environments can be achieved.

[0041] The present invention provides an industrial camera fill light source adaptive control method, which can more accurately evaluate the fill light requirements of different shooting scenes by acquiring scene perception data and performing real-time classification and environmental parameter analysis, thereby improving the accuracy of fill light control and providing a more reliable basis for image acquisition. By performing image quality evaluation on scene feature information and comparing it with preset standards, refined management of fill light requirements for different scenes is achieved, which helps to achieve on-demand fill light and avoid waste of light source resources. Control parameter analysis based on image quality matching and scene feature information can ensure that the fill light system can operate efficiently under different environmental conditions, reduce unnecessary energy consumption, and improve the overall operation efficiency of the system. By performing brightness distribution and detail feature analysis on image data and performing dynamic light source simulation, a more reasonable fill light strategy is formulated, thereby effectively utilizing light source resources and improving the stability of image quality. By comprehensively processing initial fill light control parameters, image analysis results and dynamic fill light source parameters, the fill light strategy can be flexibly adjusted according to the characteristics and demand changes of different scenes, so that the system is more adaptable to diversified industrial application scenarios, effectively solving the problem that traditional fixed fill light methods are difficult to adapt to complex environments.

[0042] In one embodiment, scene perception data of an industrial camera is obtained, and real-time classification of shooting scenes and environmental parameter analysis are performed based on the scene perception data to obtain scene feature information, including: The spectral data and three-dimensional depth information of the scene are collected synchronously through industrial cameras. During the data collection process, the spectral sensor captures the characteristics of the electromagnetic spectrum with a wavelength range of 380-780nm, and the depth sensor uses structured light or time-of-flight (ToF) technology to accurately measure the spatial position and distance of objects in the scene.

[0043] In the data integration stage, the high-order singular value decomposition (Higher-order SVD) technology is used to convert the spectral and depth data into a cross-modal correlation feature matrix. This matrix effectively reduces the data dimension and extracts key semantic information through feature decoupling and information compression. In the process of constructing the modal coupling matrix, the constraints of information theory and manifold learning are introduced to ensure the robustness of feature expression and information integrity.

[0044] The scene semantic graph maps the modal coupling matrix to a set of topologically connected semantic nodes. Each node represents a key semantic feature in the scene, and the connection weights between nodes reflect the strength of the association between features. During the semantic graph generation process, graph structure regularization constraints are introduced to enhance the structural consistency of the graph representation.

[0045] The topological feature vector is calculated through the persistent homology algorithm to calculate the Betti number sequence, which reflects the topological complexity of the scene. The persistent homology method is based on algebraic topology theory and can capture the multi-scale geometric and topological characteristics of the data and quantitatively describe the scene structure. During the calculation process, the scale parameter threshold is set to control the granularity of feature extraction.

[0046] Multi-scale fractal dimension analysis uses the theory of generalized fractal dimension to quantify the complexity of scenes from different scales and angles. The analysis process introduces self-similarity and scale invariance constraints to ensure the stability and representativeness of the complexity spectrum. The scene complexity spectrum not only reflects the spatial structural characteristics, but also reveals the dynamic change rules of the scene.

[0047] Dynamic feature analysis is based on nonlinear dynamics theory, and the stability index of the scene is calculated through phase space reconstruction and state transfer matrix. The analysis process takes into account the nonlinear coupling and chaotic characteristics of the system, and sets criteria such as Lyapunov index and entropy complexity to characterize the dynamic evolution characteristics of the scene.

[0048] Symbolization converts continuous dynamic features into discrete symbol sequences, using fuzzy clustering and information quantification methods. In the process of symbol sequence generation, information entropy and fuzzy membership constraints are introduced to ensure the information integrity and discrimination of symbolic representation.

[0049] Conditional entropy and mutual information calculations reveal the uncertainty and information complexity of the scene. The calculation process is based on the principles of probability distribution and information theory, setting information entropy thresholds and mutual information criteria to quantify the randomness and structure of the scene.

[0050] The topological feature vectors, dynamic features and uncertainty indicators are fused through nonlinear mapping to generate comprehensive scene feature information.

[0051] This embodiment realizes precise and intelligent control of the fill light source for industrial cameras through cross-modal data fusion and complex scene analysis. It can significantly improve the adaptability of the light source and can effectively meet the lighting requirements of different complex scenes. Through continuous coherence and multi-scale fractal dimension analysis, the topological structure and dynamic characteristics of the scene can be accurately captured, overcoming the limitations of traditional methods for environmental perception. By introducing symbolic processing and information theory analysis, the robustness of scene feature extraction and information utilization efficiency are significantly improved. Compared with traditional fill light solutions, this embodiment can more intelligently adjust the light source intensity and spectral characteristics, significantly improve the quality and consistency of industrial image acquisition, and provide more reliable technical support for precision industrial detection.

[0052] In one embodiment, image quality evaluation is performed on scene feature information and compared with a preset image quality standard to obtain an image quality matching degree, including: The image quality of the input scene feature image is evaluated. The brightness entropy value of the image is calculated by the entropy formula, which reflects the uniformity of the grayscale distribution of the image. According to the entropy value, a scoring mechanism of 0-100 is used to quantitatively evaluate the brightness uniformity of the image. The closer the entropy value is to the ideal uniform distribution, the higher the brightness uniformity score is.

[0053] The image is subjected to texture spectrum analysis, and the spectrum energy distribution is calculated using Fast Fourier Transform (FFT), focusing on the analysis of high, medium, and low frequency components. The detail retention score is obtained by evaluating the retention rate of high frequency components. The higher the retention rate, the more complete the image detail information is preserved, and the higher the score.

[0054] Convert the image to HSV color space, calculate the saturation histogram and average saturation value. Based on the saturation value, use a 0-100 score mechanism to evaluate the color reproduction of the image. The closer the saturation value is to the ideal range, the higher the color reproduction score.

[0055] In the comprehensive image quality scoring phase, the brightness uniformity (weight 0.3), detail retention (weight 0.4) and color reproduction (weight 0.3) are weighted and calculated according to the preset weights. In this way, a comprehensive image quality score is obtained.

[0056] Finally, the comprehensive image quality score is analyzed for matching with the preset image quality standard threshold (e.g., 85 points). The matching percentage between the score and the standard is calculated, and the relevant parameters of the fill light source are adjusted according to the matching feedback to achieve adaptive fill light control for industrial scenes.

[0057] In the specific implementation, the following scoring rules are set: Brightness uniformity score: Entropy value >4.5: 90-100 points, entropy value 3-4.5: 70-89 points, entropy value <3: 0-70 points.

[0058] Detail Preservation Rating: High frequency component retention rate>80%: 90-100 points, retention rate 60-80%: 70-89 points, Retention rate <60%: 0-70 points.

[0059] Color reproduction rating: Saturation 0.6-1.0: 90-100 points, saturation 0.3-0.6: 70-89 points, saturation <0.3: 0-70 points.

[0060] In one embodiment, the control parameter analysis is performed based on the image quality matching degree and the scene feature information to obtain the initial fill light control parameters, including: Through in-depth analysis of image quality matching and scene feature information, intelligent adaptive control of supplementary light sources is achieved. The image quality matching is evaluated based on the similarity index between the original image and the ideal reference image, and the structural similarity index (SSIM) is used as the quantitative standard of the matching. The matching value is transformed nonlinearly by the hyperbolic tangent function to enhance the sensitivity and discrimination of the features.

[0061] The scene feature information extraction adopts a multi-dimensional feature space construction strategy, and first performs high-order statistical feature extraction. The image grayscale histogram is calculated to extract high-order moment features such as skewness and kurtosis to reflect the complexity and concentration of the image grayscale distribution. At the same time, the variance and standard deviation of the pixel grayscale distribution are calculated to quantify the contrast and brightness changes of the image. Frequency domain feature extraction performs Fourier transform on the image, analyzes the spectrum energy distribution, and extracts the texture and structural features of the image. The high-order statistical features and frequency domain features are spliced ​​to form a multi-dimensional scene feature vector V, which lays the foundation for subsequent parameter analysis.

[0062] The initial fill light parameter set is generated based on the mapping relationship between the quality matching curve Q and the scene feature vector V. The fill light intensity, color temperature, strobe frequency and other parameters are determined according to predefined rules. The mapping model considers factors such as scene brightness, contrast, and texture complexity, and obtains the initial fill light parameter set P through nonlinear mapping of multidimensional feature space. For example, when the brightness variance in the scene feature vector V is large, the fill light intensity is increased; when the frequency domain characteristics show more high-frequency components, the color temperature is adjusted to improve image clarity.

[0063] Furthermore, feature points are extracted from the quality matching curve, and frequency domain feature parameters of the curve are extracted through Fourier transform. Principal component analysis is performed on the scene feature vector to construct a feature space after dimensionality reduction.

[0064] Based on the frequency domain characteristic parameters of the quality matching curve and the principal components of the scene feature vector, a multidimensional parameter mapping model is established. The parameter weights of the multidimensional parameter mapping model are calculated to determine the importance weights of each feature dimension. According to the parameter weights, the initial fill light parameter set is fitted by the least squares method to obtain the initial solution of the parameter set. The constraint condition test is performed on the initial fill light parameter set to eliminate the parameters that do not meet the preset constraint conditions.

[0065] Among them, feature points are extracted from the quality matching curve, and frequency domain feature parameters of the curve are extracted through Fourier transform, including: The quality matching curve is subjected to discrete Fourier transform to obtain the frequency domain spectrum coefficients, the amplitude and phase information of the frequency domain spectrum coefficients are extracted, and the characteristic parameter vector is constructed based on the amplitude and phase of the spectrum coefficients.

[0066] Perform principal component analysis on the scene feature vector to construct a feature space after dimensionality reduction, including: Calculate the covariance matrix of the scene feature vector, perform eigenvalue decomposition on the covariance matrix, select the main feature components according to the size of the eigenvalues, and reconstruct the feature space with the main feature components.

[0067] Parameter weight calculation based on multi-dimensional parameter mapping model, including: Construct a parameter sensitivity matrix and calculate the information gain of each parameter through information entropy.

[0068] The iterative optimization of image clarity introduces a preset evaluation function E(P). Common indicators include image entropy and gradient variance. Image entropy reflects the amount of image information, and gradient variance reflects the clarity of edge details. The gradient calculation is performed on the initial parameter set P, and the sensitivity of the evaluation function E(P) to each parameter is analyzed. Through partial derivative calculation, the optimization gradient information G is obtained to guide the direction of parameter adjustment. The gradient descent method is a commonly used optimization strategy. The update formula is P'=P-α*G, where α is the learning rate and controls the step size of each iteration.

[0069] The initial fill light parameter set is dynamically adjusted based on the optimized gradient information. During the iteration process, the image quality changes are monitored in real time, and the fill light control parameters are continuously optimized. When the gradient converges or reaches the preset number of iterations, the optimal fill light control parameter set is output. In practical applications, the nonlinear transformation coefficients, feature extraction algorithms, and optimization strategies need to be adjusted according to the characteristics of specific industrial scenarios to adapt to different image acquisition environments.

[0070] This embodiment evaluates image quality through the structural similarity index (SSIM), and combines multi-dimensional scene feature analysis with high-order statistical features and frequency domain features to achieve precise adjustment of the fill light source. Feature sensitivity is enhanced through nonlinear transformation to ensure that the dynamic adjustment of fill light parameters is more in line with actual scene requirements, thereby improving image clarity and detail performance. By adopting principal component analysis and the construction of parameter sensitivity matrix, the weight calculation of feature dimensions is made more scientific and reasonable, and parameters that do not meet the preset conditions can be effectively eliminated, thereby optimizing the fill light control strategy. An iterative optimization mechanism for image clarity is introduced, and image quality changes are monitored in real time through evaluation functions to ensure continuous adjustment and optimization of fill light parameters, and finally output the optimal parameter set. This method not only improves the adaptability and flexibility of image acquisition, but also reduces image quality fluctuations caused by environmental changes, providing a more stable and efficient solution for industrial applications.

[0071] In one embodiment, image data of an industrial camera is obtained, and brightness distribution and detail feature analysis are performed on the image data to obtain corresponding image analysis results, including: The image data undergoes multi-scale pyramid decomposition, which uses the wavelet transform algorithm to decompose the original image into frequency coefficients of different scales and directions. The decomposition process includes low-frequency and high-frequency sub-bands, where the low-frequency sub-bands retain the overall structural information of the image, and the high-frequency sub-bands capture the details and texture features of the image. For each sub-band, its energy distribution is calculated to generate a sub-band frequency domain feature map. This step helps to quantify the information characteristics of the image at different scales.

[0072] In the regional analysis stage, the sub-band frequency domain feature map is segmented by regional values, and the salient regions in the image are identified by setting thresholds and region growing algorithms. The salient region matrix marks representative image regions in the image, which may contain key targets or important texture information. Based on the salient region matrix, the image is divided into different target regions, each of which represents different image semantic content.

[0073] Local directional feature analysis is a key step in image understanding. The local directional gradient of the target area is calculated, and the local directional feature vector is constructed by analyzing the gradient direction and intensity of the pixel points. This step can capture the directional characteristics of the texture in the image and provide a basis for subsequent texture complexity evaluation.

[0074] In order to quantify the texture complexity of the image, the pixel direction consistency of the local directional features is calculated to generate a detail descriptor. The detail descriptor maps the directional features to discrete complexity levels through clustering analysis methods (such as K-means clustering) to obtain the texture complexity index of the image. This index reflects the uniformity and structural complexity of the image texture.

[0075] In the matrix construction stage, the joint distribution matrix is ​​constructed by combining the sub-band frequency domain feature map and the texture complexity index. The matrix construction follows the principle of maximizing information entropy and captures the multi-dimensional information characteristics of the image through feature fusion. The joint distribution matrix is ​​vector-feature decomposed to extract the main information feature vector of the image.

[0076] Image quality is scored based on the image matrix vector and the joint distribution matrix. Considering multiple dimensions such as brightness distribution, texture complexity, and frequency domain features, a comprehensive image quality evaluation index is output.

[0077] This embodiment significantly improves the image quality of industrial cameras under complex lighting conditions through multi-scale image analysis technology. The use of pyramid decomposition and sub-band frequency domain feature analysis can accurately capture the details and texture information in the image, thereby effectively identifying salient areas. This process ensures that under different ambient lighting conditions, the industrial camera can automatically adjust the supplementary light source, optimize the image acquisition effect, and improve the clarity and recognizability of the image. The calculation of local directional feature analysis and detail descriptors makes the assessment of image texture complexity more accurate, thereby improving the reliability of image quality scoring. By constructing a joint distribution matrix and comprehensively considering frequency domain features and texture complexity, a comprehensive understanding of image information is achieved, providing a scientific basis for the adaptive adjustment of supplementary light sources.

[0078] In one embodiment, dynamic light source simulation is performed on scene feature information based on the image analysis result to obtain corresponding dynamic supplementary light source parameters, including: Perform a comprehensive lighting uniformity assessment on the input image, using algorithms such as pixel brightness standard deviation and histogram comparison to calculate the lighting differences in different areas of the image. When the lighting uniformity is detected to be lower than the preset threshold (for example, 0.6), the virtual adjustment process of the light source position will be triggered.

[0079] The generation of virtual light source position parameters is based on the scene geometry and illumination distribution relationship, and multiple sets of candidate light source positions are constructed through a spatial mapping algorithm. Each set of position parameters is combined with scene depth information and occlusion relationship, and the illumination distribution data is calculated using the Monte Carlo ray tracing algorithm. The ray tracing process takes into account the reflection characteristics of the surface material, simulates the propagation and attenuation characteristics of light on different material surfaces, and generates a high-precision illumination distribution map.

[0080] Surface reflection characteristic analysis combines image texture and material recognition algorithms to construct a reflectivity distribution map of each object surface in the scene. The analysis process includes texture gradient calculation, material classification and reflectivity parameter extraction to identify the optical characteristics of different materials. Based on the reflectivity distribution map, the original illumination distribution data is accurately corrected to compensate for the complex influence of materials on light propagation.

[0081] Information theory is introduced into the light intensity optimization calculation. Through entropy analysis and visual saliency algorithm, the fidelity of image information and visual comfort under different light intensities are evaluated. The optimization goal is to maximize the contrast and clarity of the image while maintaining image details. The calculation process constructs a multi-dimensional optimization objective function, comprehensively considering indicators such as brightness uniformity, color saturation, and texture detail retention.

[0082] Candidate fill-light solutions are comprehensively evaluated through an image quality prediction model, and the image quality evaluation network based on deep learning integrates objective indicators such as peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) and subjective perception scores. The evaluation algorithm generates multi-dimensional quality prediction indicators for each group of fill-light solutions, and uses a weighted sorting strategy to screen out the optimal fill-light solution.

[0083] The dynamic fill light source parameter generation process involves smooth mapping of light source position and light intensity, and the continuous transition of parameters is achieved through interpolation algorithm to avoid sudden changes in lighting. Parameter conversion takes into account the dynamic changes of the scene and the perception characteristics of the human eye to ensure the natural smoothness of the fill light process. The entire method realizes intelligent and adaptive light source simulation based on image analysis, significantly improving the visual quality and information readability of the image.

[0084] This embodiment solves the problem of local overexposure or dark areas caused by the fixed position of the light source in the traditional fill-light solution by evaluating the uniformity of the illumination of the image and adjusting the position of the virtual light source. The reflectivity distribution map is constructed by using a micro-surface reflection model and a material recognition algorithm, which overcomes the limitation of conventional fill-light technology that ignores material characteristics and makes the fill-light effect closer to the real scene. The light intensity parameters are optimized by information theory and visual perception models, which improves the overall visual comfort of the picture while ensuring the image details. The image quality prediction model based on deep learning evaluates and screens the fill-light solution, avoiding the blindness of the selection of fill-light parameters in the traditional method. Finally, the dynamic adjustment of the illumination parameters is realized through a smooth mapping algorithm, which eliminates the sudden change of light intensity during the fill-light process and improves the naturalness and coherence of the fill-light effect. The overall solution significantly improves the intelligence level and adaptability of scene fill-light and enhances the image quality.

[0085] In one embodiment, the initial fill light control parameters, the image analysis results and the dynamic fill light source parameters are comprehensively processed to obtain a fill light source adjustment strategy, including: When performing fractal dimension analysis on the initial fill light control parameters, the fill light complexity index is obtained by calculating the box count dimension and information dimension of the fill light parameters. The box count dimension reflects the spatial distribution characteristics of the fill light parameters, and the information dimension characterizes the information entropy distribution characteristics of the parameters. In the specific calculation, the fill light control parameter space is divided into several grids of equal size, the number of parameter points in each grid is counted, and the dimension value is calculated using fractal theory, and finally an index value reflecting the complexity of fill light control is obtained.

[0086] In the process of constructing the optical flow field, the pixel displacement vectors between adjacent frames are calculated based on the image analysis results, the Horn-Schunck algorithm is used to establish the optical flow constraint equation, and the motion vector field of the entire scene is obtained by solving the variational method. The optical flow field calculation must meet the assumptions of image grayscale constancy and motion continuity, and finally generate an optical flow field map that represents the dynamic characteristics of the scene.

[0087] When performing tensor decomposition on the dynamic supplementary light source parameters, the parameters such as brightness, color temperature, and illumination angle of the supplementary light source are organized into a third-order tensor, which is decomposed into the product of the core tensor and the factor matrix. During the decomposition process, an appropriate rank parameter is set to ensure that the extracted features have good expressive power, and finally a feature representation that reflects the multidimensional attributes of the supplementary light source is obtained.

[0088] In the topology construction phase, the number of network nodes is determined by the fill light complexity index, and the connection relationship between nodes is determined by the scene dynamic optical flow field diagram to build a fill light control network with adaptive capabilities. The network structure must meet the requirements of connectivity and stability, and the connection weights between nodes are dynamically adjusted according to the gradient information of the optical flow field.

[0089] The feature mapping process uses a nonlinear projection method to map the multimodal light source features to the network nodes. The activation state of the node reflects the value of the corresponding fill light parameter. The mapping function uses a nonlinear function to ensure the smoothness of the feature projection, thereby obtaining a preliminary fill light control strategy.

[0090] In the optimization of the pulse sequence of the fill light source control, the pulse waveform and timing parameters are designed based on the preliminary fill light strategy. The optimization goals include minimizing energy consumption, ensuring lighting uniformity and avoiding stroboscopic effects. The optimal pulse sequence parameters are solved by genetic algorithm, and finally a complete fill light source adjustment strategy is formed.

[0091] This embodiment achieves accurate quantification of the spatial distribution characteristics of the fill light parameters by performing fractal dimension analysis on the initial fill light control parameters, thereby effectively improving the accuracy of the fill light control. By constructing a dynamic optical flow field of the scene, the motion characteristics in the scene are accurately captured, providing reliable dynamic information support for the formulation of the fill light strategy. The adaptive fill light network constructed based on the fill light complexity index and the dynamic characteristics of the scene has a strong ability to adapt to the environment and can flexibly adjust the fill light strategy according to scene changes. Through nonlinear feature mapping and pulse sequence optimization, energy consumption is minimized while ensuring the lighting effect, effectively avoiding the influence of the stroboscopic effect on the imaging quality. The entire control method fully considers multiple key factors in the fill light process, and significantly improves the fill light control effect and image acquisition quality of industrial cameras in complex scenes.

[0092] In one embodiment, the multimodal light source characteristics are mapped to the adaptive fill light network structure to obtain a preliminary fill light strategy, including: When performing eigenvector decomposition on multimodal light source features, the principal component analysis method is used to decompose the light source feature matrix into the product of the eigenvector matrix and the eigenvalue diagonal matrix. The eigenvalues ​​are arranged from large to small, and the eigenvectors with a cumulative contribution rate of 95% are selected as the principal components of the light source features.

[0093] When assigning weights to the adaptive fill light network structure, the correlation coefficient between nodes is calculated based on the principal component of the light source characteristics, and the correlation coefficient is used as the weight value of the connecting edge. The weight assignment must meet the network connectivity requirements, and there must be at least one valid path between any two nodes.

[0094] In the spectrum analysis stage, the weighted light supplement network is spectrally clustered to extract the topological features of the network. The network structure eigenvector is obtained by calculating the eigenvalue and eigenvector of the network's Laplace matrix. The eigenvector contains topological information such as the connectivity and clustering coefficient of the network.

[0095] The acquisition of the fusion feature tensor is achieved by performing a tensor product operation on the network structure feature vector and the main component of the light source feature. The tensor product operation maintains the correlation between the two types of features and forms a high-dimensional feature representation. During the operation, it is necessary to ensure that the tensor dimensions match.

[0096] When performing nonlinear transformation on the fused feature tensor, the ReLU activation function is used to realize nonlinear mapping of features and generate a candidate set of fill-in light strategies. Nonlinear transformation enhances the expressive power of features and makes the fill-in light strategy more adaptable.

[0097] The selection of fill-light strategies is based on preset sorting criteria, including indicators such as illumination uniformity, energy efficiency, and spot distribution. The candidate strategies are scored for multiple objectives, and the strategies with the top 30% comprehensive scores are selected as the selection fill-light strategy set.

[0098] In the integration processing stage, the strategies in the selected light-filling strategy set are weighted averaged, and the weight coefficient is determined by the score value of each strategy. The integrated strategy must meet the constraints such as light intensity and energy consumption to form the final preliminary light-filling strategy.

[0099] This embodiment realizes the effective extraction of light source features and the optimal configuration of network structure by decomposing the feature vectors of multimodal light source features and mapping them to the adaptive fill-light network structure. The principal components of light source features are extracted based on the principal component analysis method, which retains the key feature information and reduces data redundancy. The spectral clustering method is used to extract network topology features, which effectively captures the structural characteristics of the network. Feature fusion is achieved through tensor product operations, which enhances the feature expression ability and makes the fill-light strategy more adaptable. In the process of strategy screening, a multi-objective scoring mechanism is adopted to ensure the illumination uniformity and energy efficiency of the selected fill-light strategy. Finally, the strategies are integrated by weighted average, and a fill-light solution that takes into account both illumination effects and system performance is formed on the premise of meeting the constraints, thereby improving the imaging quality and system operation efficiency of industrial cameras in different scenarios.

[0100] Reference Figure 2 As shown, the present invention also provides an industrial camera supplementary light source adaptive control device, which is applied to any one of the above industrial camera supplementary light source adaptive control methods, including: The acquisition module is used to obtain scene perception data of industrial cameras, perform real-time classification of shooting scenes and environmental parameter analysis based on the scene perception data, and obtain scene feature information; An analysis module is used to evaluate the image quality of scene feature information and compare it with a preset image quality standard to obtain an image quality matching degree; An association module, which is used to analyze control parameters based on image quality matching and scene feature information to obtain initial fill light control parameters; A processing module is used to obtain image data from an industrial camera, perform brightness distribution and detail feature analysis on the image data, and obtain corresponding image analysis results; A control module, which is used to perform dynamic light source simulation on scene feature information based on image analysis results to obtain corresponding dynamic supplementary light source parameters; The execution module is used to comprehensively process the initial fill light control parameters, image analysis results and dynamic fill light source parameters to obtain a fill light source adjustment strategy.

[0101] The present invention provides an industrial camera fill light source adaptive control device, which can more accurately evaluate the fill light requirements of different shooting scenes by acquiring scene perception data and performing real-time classification and environmental parameter analysis, thereby improving the accuracy of fill light control and providing a more reliable basis for image acquisition. By performing image quality evaluation on scene feature information and comparing it with preset standards, refined management of fill light requirements for different scenes is achieved, which helps to achieve on-demand fill light and avoid waste of light source resources. Control parameter analysis based on image quality matching and scene feature information can ensure that the fill light system can operate efficiently under different environmental conditions, reduce unnecessary energy consumption, and improve the overall operation efficiency of the system. By performing brightness distribution and detail feature analysis on image data and performing dynamic light source simulation, a more reasonable fill light strategy is formulated, thereby effectively utilizing light source resources and improving the stability of image quality. By comprehensively processing initial fill light control parameters, image analysis results and dynamic fill light source parameters, the fill light strategy can be flexibly adjusted according to the characteristics and demand changes of different scenes, so that the system is more adaptable to diversified industrial application scenarios, effectively solving the problem that traditional fixed fill light methods are difficult to adapt to complex environments.

[0102] It should be noted that technicians in the relevant technical field can clearly understand that for the convenience and conciseness of description, the specific working process of the system and each module described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0103] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. An adaptive control method for supplementary light source of an industrial camera, characterized in that: include: Acquire scene perception data of the industrial camera, perform real-time classification of the shooting scene and environmental parameter analysis based on the scene perception data, and obtain scene feature information; Performing image quality evaluation on the scene feature information and comparing it with a preset image quality standard to obtain an image quality matching degree; Performing control parameter analysis based on the image quality matching degree and the scene feature information to obtain initial fill light control parameters; Acquire image data of the industrial camera, perform brightness distribution and detail feature analysis on the image data, and obtain corresponding image analysis results; Performing dynamic light source simulation on the scene feature information based on the image analysis result to obtain corresponding dynamic supplementary light source parameters; The initial fill light control parameter, the image analysis result and the dynamic fill light source parameter are comprehensively processed to obtain a fill light source adjustment strategy.

2. The method for adaptively controlling supplementary light sources for industrial cameras according to claim 1, characterized in that: The step of acquiring scene perception data of an industrial camera, and performing real-time classification of shooting scenes and environmental parameter analysis based on the scene perception data to obtain scene feature information includes: Acquire the spectral data and depth data collected by the industrial camera, and integrate them to obtain the scene perception data; Performing high-order singular value decomposition on the scene perception data and extracting cross-modal correlation features to obtain a modal coupling matrix; Based on the modal coupling matrix, a scene is constructed to obtain a scene semantic graph; Performing Betti number sequence calculation on the scene semantic graph according to a preset continuous homology algorithm to obtain a scene topology feature vector; Performing multi-scale fractal dimension analysis based on the scene topology feature vector to obtain a scene complexity spectrum; Performing dynamic stability characteristic analysis based on the scene complexity spectrum to obtain scene dynamic characteristics; Symbolize the dynamic features of the scene to generate a symbol sequence of the scene; Calculating conditional entropy and mutual information based on the symbol sequence to obtain a scene uncertainty index; The scene topology feature vector, the scene dynamic feature and the scene uncertainty index are subjected to nonlinear mapping fusion to obtain the scene feature information.

3. The method for adaptively controlling supplementary light sources for industrial cameras according to claim 1, characterized in that: The performing image quality assessment on the scene feature information and comparing it with a preset image quality standard to obtain an image quality matching degree includes: Calculating the brightness entropy value of the scene feature information to obtain a scene brightness entropy value; Performing brightness uniformity evaluation on the scene feature information according to the scene brightness entropy value to obtain a brightness uniformity score; Performing texture spectrum analysis on the scene feature information to obtain texture spectrum features; Performing detail retention evaluation on the scene feature information according to the texture spectrum feature to obtain a detail retention score; Performing color saturation calculation on the scene feature information to obtain a color saturation value; Performing a color restoration evaluation on the scene feature information according to the color saturation value to obtain a color restoration score; Comprehensively calculating the brightness uniformity score, the detail retention score, and the color reproduction score to obtain a comprehensive image quality score; A matching degree analysis is performed between the comprehensive image quality score and the image quality standard to obtain an image quality matching degree.

4. The method for adaptively controlling supplementary light sources for industrial cameras according to claim 1, characterized in that: The performing control parameter analysis based on the image quality matching degree and the scene feature information to obtain initial fill light control parameters includes: Performing a nonlinear transformation of a hyperbolic tangent function on the image quality matching degree to obtain a quality matching curve; Extracting high-order statistical features and frequency domain features from the scene feature information, and constructing a multi-dimensional feature space to obtain a scene feature vector; Performing parameter analysis on the quality matching curve and the scene feature vector to obtain an initial fill light parameter set; Iteratively calculating the initial fill light parameter set according to a preset image clarity evaluation function to obtain optimized gradient information; The initial fill light parameter set is dynamically adjusted according to the optimization gradient information to obtain the initial fill light control parameters.

5. The method for adaptively controlling supplementary light sources for industrial cameras according to claim 1, characterized in that: The acquiring of the image data of the industrial camera, performing brightness distribution and detail feature analysis on the image data, and obtaining corresponding image analysis results includes: Performing multi-scale pyramid decomposition on the image data to obtain multi-scale frequency coefficients; Perform energy distribution analysis of each sub-band based on the multi-scale frequency coefficients to obtain a sub-band frequency domain feature map; Performing region value segmentation on the sub-band frequency domain feature map to obtain a significant region matrix; Divide the image data into regions according to the salient region matrix to obtain a target region; Performing local directional feature analysis on the target area to obtain local directional features; Calculating the direction consistency of regional pixels based on the local direction features to obtain a detail descriptor; Performing cluster analysis on the detail descriptors to obtain an image texture complexity index; Performing matrix construction based on the sub-band frequency domain feature map and the texture complexity index to obtain a constructed joint distribution matrix; Performing vector eigendecomposition on the joint distribution matrix to obtain an image matrix vector; An image quality score is performed based on the image matrix vector and the joint distribution matrix to obtain the image analysis result.

6. The method for adaptively controlling supplementary light sources for industrial cameras according to claim 1, characterized in that: The performing dynamic light source simulation on the scene feature information based on the image analysis result to obtain corresponding dynamic supplementary light source parameters includes: Performing illumination non-uniformity evaluation on the image analysis result to obtain illumination uniformity information; Virtually adjusting the light source position of the scene feature information according to the illumination uniformity information to obtain multiple groups of virtual light source position parameters; Performing ray tracing calculation on each group of virtual light source position parameters to obtain corresponding illumination distribution data; Analyzing the object surface reflection characteristics of the scene feature information according to the illumination distribution data to obtain a surface reflectivity distribution map; Correcting the illumination distribution data based on the surface reflectivity distribution map to obtain corrected illumination distribution data; Performing light intensity optimization calculation on the corrected light distribution data to obtain optimized light intensity parameters; Generate multiple groups of candidate fill light solutions according to the optimized light intensity parameters and the virtual light source position parameters; Performing image quality prediction and evaluation on each group of candidate fill-light solutions to obtain corresponding image quality prediction indicators; Sorting and screening the candidate fill light solutions according to the image quality prediction index to obtain a screened fill light solution; The light source position and light intensity parameters are dynamically converted for the screened fill light scheme to obtain the dynamic fill light source parameters.

7. The method for adaptively controlling supplementary light sources for industrial cameras according to claim 1, characterized in that: The comprehensive processing of the initial fill light control parameter, the image analysis result and the dynamic fill light source parameter to obtain a fill light source adjustment strategy includes: Performing fractal dimension analysis on the initial fill light control parameters to obtain a fill light complexity index; Constructing an optical flow field according to the image analysis results to obtain a scene dynamic optical flow field map; Performing tensor decomposition on the dynamic supplementary light source parameters to obtain multimodal light source features; A topological structure is constructed based on the fill light complexity index and the scene dynamic optical flow field map to obtain an adaptive fill light network structure; Mapping the multimodal light source characteristics onto the adaptive fill light network structure to obtain a preliminary fill light strategy; The fill light source control pulse sequence is optimized based on the preliminary fill light strategy to obtain the fill light source adjustment strategy.

8. The method for adaptively controlling supplementary light sources for industrial cameras according to claim 7, characterized in that: The step of mapping the multimodal light source characteristics onto the adaptive fill light network structure to obtain a preliminary fill light strategy includes: Performing feature vector decomposition on the multimodal light source features to obtain main components of light source features; Assigning node weights to the adaptive fill-light network structure according to the main components of the light source characteristics to obtain a weighted fill-light network; Performing graph analysis on the weighted light supplement network and extracting network topology features to obtain a network structure feature vector; Performing a tensor product operation on the network structure feature vector and the light source feature principal component to obtain a fused feature tensor; Performing nonlinear transformation on the fused feature tensor to obtain a candidate set of fill light strategies; The fill light strategy candidate set is screened according to a preset fill light sorting standard to obtain a corresponding screened fill light strategy set; The screened fill-light strategy set is integrated to obtain the preliminary fill-light strategy.

9. An adaptive control device for supplementary light source of an industrial camera, characterized in that: The method for adaptively controlling supplementary light sources for industrial cameras applied to any one of claims 1 to 8 above comprises: An acquisition module, which is used to obtain scene perception data of an industrial camera, and to perform real-time classification of shooting scenes and environmental parameter analysis based on the scene perception data to obtain scene feature information; An analysis module, the analysis module is used to evaluate the image quality of the scene feature information and compare it with a preset image quality standard to obtain an image quality matching degree; An association module, the association module is used to perform control parameter analysis based on the image quality matching degree and the scene feature information to obtain initial fill light control parameters; A processing module, the processing module is used to obtain image data of the industrial camera, perform brightness distribution and detail feature analysis on the image data, and obtain corresponding image analysis results; A control module, the control module is used to perform dynamic light source simulation on the scene feature information based on the image analysis result to obtain corresponding dynamic supplementary light source parameters; An execution module is used to comprehensively process the initial fill light control parameters, the image analysis results and the dynamic fill light source parameters to obtain a fill light source adjustment strategy.

Images