Product color matching database construction method and system based on data collection

Abstract

The invention provides a product color matching database construction method and system based on data collection, and relates to the technical field of electrical digital data processing.The method comprises the steps that multi-dimensional optical parameters are collected through a distributed spectrum sensor array, chromaticity offset is dynamically corrected based on an environment adaptation coefficient, and a color matching database is obtained; a sensing weight matrix is calculated by combining material features and observation angles, multi-dimensional data fusion is realized by utilizing high-order tensor operation, a hierarchical color matching database is constructed by adopting an improved OPTICS algorithm, and weight distribution is dynamically optimized through a back propagation mechanism; the system comprises a high-precision spectrum sensor, a GPU / FPGA heterogeneous computing architecture, a tensor fusion coprocessor and a user behavior analysis engine. The problem of perception deviation of a traditional linear model is solved, cross-material and multi-scene color adaptive matching is achieved through a spectrum compensation and dynamic clustering cooperation mechanism, virtual reality and other emerging field applications are supported, and high precision, high robustness and real-time processing capacity are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of electrical digital data processing, and particularly to a method and system for constructing a product color matching database based on data collection. Background Art

[0002] With the rapid development of digital technology, product color matching design has gradually shifted from traditional manual experience-driven to data-driven intelligent models. However, existing color matching scheme construction technologies still have significant limitations, especially in terms of dynamic environment adaptability, cross-material generalization ability, and database self-optimization mechanism, which urgently need to be broken through. Traditional methods mostly rely on static color databases and linear matching algorithms, and it is difficult to cope with the influence of complex lighting conditions, observation angle changes, and material diversity on color perception. For example, existing color matching systems usually perform color matching under fixed environmental parameters. When the light intensity or color temperature of the actual application scenario deviates, it is easy to cause a significant deviation between the color matching result and the expectation, seriously restricting the reliability of cross-scene applications.

[0003] In current technologies, the construction of color matching databases is mostly based on single-dimensional chromaticity parameters (such as RGB or LAB values), lacking quantitative modeling of human visual perception characteristics, resulting in a difference between the color matching scheme and the actual sensory experience. In addition, existing clustering algorithms (such as K-means) are difficult to effectively identify the semantic relationship between the main color and the secondary color when processing high-dimensional chromaticity data, and database updates rely on manual intervention and cannot be dynamically optimized according to user behavior data. At the hardware implementation level, traditional systems are limited by the rigid design of the computing architecture and are difficult to meet the real-time processing requirements, especially facing efficiency bottlenecks in multi-source data fusion and high-order tensor operations.

[0004] In response to the above problems, some improvement schemes have been proposed in the prior art, such as introducing an ambient light compensation algorithm or a weight adjustment method based on texture analysis. However, these schemes are often limited to a single technical dimension, lack systematic integration, and do not solve core problems such as poor cross-material adaptability and lagging database updates. In addition, existing systems mostly adopt a centralized computing architecture, which is difficult to support distributed deployment and elastic expansion, restricting the application potential in emerging fields such as industrial Internet of Things and virtual reality. Therefore, there is an urgent need for a systematic solution that integrates multi-source data perception, dynamic environment adaptation, and intelligent decision optimization to break through the bottleneck of existing technologies and achieve high-precision and adaptive construction and maintenance of product color matching databases. Summary of the Invention

[0005] In order to solve the technical problems in the prior art, such as color matching deviation caused by dynamic light and observation angle changes, insufficient cross-material scene adaptability, database updates relying on manual intervention, low clustering efficiency of high-dimensional chromaticity data, and limited real-time processing ability of the hardware architecture, the present invention provides a method and system for constructing a product color matching database based on data collection.

[0006] The technical solution provided by the present invention is as follows:

[0007] In the first aspect:

[0008] A method for constructing a product color matching database based on data collection provided by the present invention includes:

[0009] S1. Multi-source data collection stage: Collect the surface reflection spectrum data of the target object through a distributed sensor array, and simultaneously obtain the ambient light parameters and the observation angle parameters, wherein the surface reflection spectrum data includes the reflectance values at intervals of 5 nm within the range of 380 - 780 nm;

[0010] S2. Feature dimension conversion stage: Convert the collected original spectrum data into three-dimensional coordinates of a chromaticity space that conforms to human visual perception, including the brightness dimension, the hue dimension, and the saturation dimension, and generate corresponding spectral feature vectors;

[0011] S3. Environment adaptation correction stage: Calculate the environment adaptation coefficient according to the illuminance value and the color temperature value in the ambient light parameters, and this coefficient includes a spectral offset compensation factor and a brightness dynamic compensation factor;

[0012] S4. Perceptual weight calculation stage: Based on the material type and the surface texture characteristics of the target object, and in combination with the observation angle parameters, calculate the color perception weight matrix, and this matrix includes the visual saliency factors of each chromaticity channel;

[0013] S5. Multi-dimensional data fusion stage: Perform a tensor product operation on the chromaticity coordinates after environment adaptation correction and the color perception weight matrix to generate a standardized color matching feature vector;

[0014] S6. Intelligent clustering analysis stage: Use an improved density clustering algorithm to classify the standardized color matching feature vectors, and establish a color matching database with a hierarchical structure, where the top-level nodes store the main color features and the sub-nodes store the auxiliary color features;

[0015] S7. Dynamic optimization and update stage: According to the preference parameters and the usage frequency parameters in the user feedback data, dynamically adjust the weight distribution of each node in the color matching database through the backpropagation algorithm.

[0016] Further, in the S3, it includes:

[0017] The calculation formula for the spectral offset compensation factor δ is:

[0018]

[0019] Among them, ΔT is the absolute temperature difference value between the ambient color temperature and the standard light source D65, with the unit of K; E is the current ambient illuminance value, with the unit of lx; E0 is the reference illuminance reference value of 2000 lx; CRI is the color rendering index of the ambient light source; k1 = 0.0032 and k2 = 0.15 are empirical coefficients.

[0020] S302. The calculation formula for the brightness dynamic compensation factor η is:

[0021]

[0022] Among them, L is the measured brightness value under the current ambient illuminance, and L max is the maximum reflection brightness of the material under the standard light source, and γ is the attenuation coefficient related to the material type, taking 0.12 for smooth surfaces and 0.08 for rough surfaces.

[0023] S303. The environmental adaptation coefficient α is finally obtained by weighted summation of δ and η: α = 0.6δ + 0.4η.

[0024] Furthermore, the specific content of S4 includes:

[0025] The calculation method of the color perception weight matrix includes the following steps:

[0026] S401. Extract the frequency domain features of the object surface texture through a convolutional neural network, and calculate the texture complexity index TCI (Texture Complexity Index):

[0027]

[0028] Among them, freq band (i) is the central frequency value of the i-th frequency band, and energy ratio (i) is the energy proportion of the corresponding frequency band.

[0029] S402. Determine the basic weight vector w base = [w L , w a , w b according to the material type, where:

[0030] For metal materials: w L = 0.4, w a = 0.35, w b = 0.25

[0031] For polymer materials: wL = 0.3, w a = 0.4, w b = 0.3

[0032] For natural materials: w L= 0.25, w a = 0.3, w b = 0.45;

[0033] S403, Introduce the correction function of the observation angle:

[0034] f(θ) = 1 + 0.5·sin(2θ) - 0.2·cos(θ)

[0035] The final color perception weight matrix w final = w base ·(1 + TCI)·f(θ).

[0036] Furthermore, the S6 specifically includes:

[0037] The improved density clustering algorithm adopts an improved scheme based on the OPTICS (Ordering Points To Identify the Clustering Structure) algorithm, specifically including:

[0038] S601, Introduce the adaptive core distance parameter:

[0039] core dist (ε) = μ·d mean +(1 - μ)·d meanian

[0040] where d mean is the average value of the k-nearest neighbor distances of the sample point, d meanian is the median of the k-nearest neighbor distances of the sample point, and μ = 0.7 is the mixing coefficient;

[0041] S602, Introduce the color similarity constraint term in the reachability distance calculation:

[0042] reach dist (x, y) = max{core dist (x), d(x, y)·S(x, y)}

[0043] where S(x, y) = exp(-β·ΔE 00 (x, y)) is the color difference similarity function, and ΔE 00 (x, y) is the color difference value calculated using the CIE2000 color difference formula, and β = 0.05 is the adjustment coefficient.

[0044] Furthermore, the S6 further includes:

[0045] The method for generating the middle-level structure includes:

[0046] S603, Calculate the main color feature extraction threshold:

[0047]

[0048] Among them, is the variance of the hue dimension, ρ ab is the covariance coefficient of hue and saturation, and ΔS is the saturation gradient change rate;

[0049] S604. Define the hue dispersion index HDI (Hue Dispersion Index):

[0051]

[0052] Among them, h i is the hue value of the clustering sample, and μ h is the average hue value;

[0053] S605. The calculation formula for the saturation gradient SG (Saturation Gradient) is:

[0054]

[0055] Among them, s i is the sorted saturation sequence, and σ s is the saturation standard deviation;

[0056] S606. When HDI > 2.5 and SG < 0.8, start the sub-node splitting mechanism, and use the improved K-means++ algorithm to generate the auxiliary color feature clusters.

[0057] Furthermore, the dynamic optimization update in S7 specifically includes:

[0058] S701. Define the weight adjustment factor ω:

[0059] ω = λ · (1 - exp(-τ · f p )) + (1 - λ) · tanh(v · f u )

[0060] Among them, f p is the preference score value, with a range of 0 - 10, f u is the usage frequency, in units of times / month, λ = 0.65 is the balance coefficient, and τ = 0.2, v = 0.15 are the attenuation coefficients;

[0061] S702. Construct the backpropagation objective function:

[0062]

[0063] Among them, F i is the original feature vector, F′ iIt is the optimized feature vector, ΔH i It is the hue adjustment amount, ΔS i It is the saturation adjustment amount, ρ = 0.3 is the regularization coefficient;

[0064] S703. Iteratively update the weights of the database nodes through the gradient descent algorithm until the objective function converges.

[0065] Second aspect:

[0066] A product color matching database construction system based on data collection provided by the present invention includes:

[0067] A multi-source data acquisition module, configured with a high-precision spectral sensor array and a six-degree-of-freedom attitude sensor, for synchronously acquiring the reflected data of the object surface and the spatial orientation information;

[0068] A feature conversion engine, including a GPU-accelerated chromaticity space converter, which performs real-time conversion from CIE LAB to CIECAM02 visual perception space;

[0069] An environment adaptation processor, integrated with a programmable light compensation unit, for calculating the environment adaptation coefficient;

[0070] A perception weight generator, built-in with a material feature library and a texture analysis algorithm, for implementing the calculation of the color perception weight matrix;

[0071] A multi-dimensional fusion core, equipped with a tensor operation coprocessor, for realizing the non-linear fusion of feature vectors;

[0072] An intelligent clustering host, equipped with a dedicated computing card with an improved density clustering algorithm;

[0073] A dynamic optimization engine, including a hybrid computing architecture of a user behavior analyzer and a backpropagation optimizer.

[0074] Further, the multi-dimensional fusion core specifically includes:

[0075] A feature orthogonalization unit, which uses the Gram-Schmidt orthogonalization method to eliminate the correlation between chromaticity dimensions;

[0076] A tensor product calculation unit, supporting fourth-order tensor operations, and executing the formula:

[0077] T ijk = ∑ m (A im ·B mjk ) + C ijk ·η

[0078] where A is the environment adaptation coefficient matrix, B is the perception weight tensor, C is the base tensor, and η is the brightness compensation factor;

[0079] The normalization output unit processes the combination of the hyperbolic tangent function and L2 normalization, and outputs a standardized color matching feature vector.

[0080] Furthermore, the environment adaptation processor specifically includes:

[0081] The spectral shift compensation unit calculates the δ value in real time, and the built-in DSP chip executes fast operations;

[0082] The dynamic brightness compensation unit is configured with a material reflectivity database and calculates the η value according to the input ratio;

[0083] The fusion output unit realizes the hardware-level calculation of α = 0.6δ + 0.4η through a weighted summation circuit, and the output delay is less than 2 μs.

[0084] Furthermore, the intelligent clustering host specifically includes:

[0085] The core distance calculation module realizes the parallel calculation of core dist (ε) using FPGA;

[0086] The reachable distance optimizer integrates an ASIC chip dedicated to color difference calculation and supports the hardware acceleration of the CIE2000 color difference formula;

[0087] The clustering structure generator is configured with a hierarchical relationship construction algorithm and executes real-time monitoring of HDI and SG indicators and sub-node splitting control;

[0088] The visualization interface includes a three-dimensional chromaticity space rendering engine and supports the interactive topological display of the color matching database.

[0089] The beneficial effects brought by the technical solution provided by the present invention at least include:

[0090] (1) In the present invention, through the dual technical means of multi-source data fusion and environment adaptation correction, the problem of color consistency across media and environments is innovatively solved. The environment adaptation coefficient dynamically compensates for the influence of different lighting conditions on color perception. Combining the perception weight calculation of material characteristics and observation angles, the color matching scheme has self-adaptability. The tensor operation in multi-dimensional data fusion constructs a non-linear mapping relationship in the high-dimensional chromaticity space, breaking through the limitations of traditional linear models. The combination of the hierarchical clustering structure and the dynamic optimization mechanism realizes the intelligent evolution ability of the color matching database, significantly improving the accuracy and generalization of color matching, and providing a universal color matching solution for multiple fields;

[0091] (2) In the present invention, the system adopts an architecture that deeply integrates perceptual computing and machine learning to achieve full-process intelligence from data collection to optimization decision-making. The collaborative work of the environment adaptation processor and the perceptual weight generator simulates the color adaptation mechanism of the human visual system, making the color matching result more in line with subjective perception. The improved density clustering algorithm combined with the HDI / SG index system can autonomously identify the association relationship between the main color and the auxiliary color, construct a color matching database with semantic levels. The dynamic optimization engine establishes a closed-loop learning system through user feedback data, enabling the database to continuously evolve and fit the personalized needs of users, greatly reducing the dependence on manual parameter adjustment and improving the color matching decision-making efficiency;

[0092] (3) In the present invention, the heterogeneous computing architecture design at the hardware level effectively balances the processing speed and energy efficiency ratio. Through the collaborative acceleration of GPU / FPGA / ASIC, it meets the real-time requirements. The modular design supports distributed deployment and elastic expansion, and can adapt to the multi-scenario requirements from embedded devices to cloud servers. The standardized interface and cross-platform compatibility design reduce the system integration complexity and facilitate docking with existing color management systems. The environment robustness enhancement technology (such as dynamic calibration, anti-interference filtering) ensures stable operation under complex conditions in industrial fields, providing reliable technical support for emerging fields such as intelligent manufacturing and digital twin. BRIEF DESCRIPTION OF THE DRAWINGS

[0093] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0094] Figure 1 It is a schematic flow chart of a method for constructing a product color matching database based on data collection provided by an embodiment of the present invention;

[0095] Figure 2 It is a schematic structural diagram of a system for constructing a product color matching database based on data collection provided by an embodiment of the present invention;

[0096] Figure 3 It is a schematic diagram of the data flow of a system for constructing a product color matching database based on data collection provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0097] The following will describe the technical solutions in the present invention with reference to the drawings.

[0098] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to give examples, illustrations or explanations. Any embodiment or design solution described as an "example" in the present invention should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Rather, the use of the word "example" is intended to present concepts in a specific manner. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one of the two can be selected.

[0099] In the embodiments of the present invention, "image" and "picture" can sometimes be used interchangeably. It should be noted that when the difference is not emphasized, the meanings they express are the same. "(of)", "corresponding", and "corresponding" can sometimes be used interchangeably. It should be noted that when the difference is not emphasized, the meanings they express are the same.

[0100] In the embodiments of the present invention, sometimes a subscript such as W1 may be miswritten as a non-subscript form such as W1. When the difference is not emphasized, the meanings they express are the same.

[0101] To make the technical problems, technical solutions and advantages to be solved by the present invention clearer, the following will be described in detail with reference to the drawings and specific embodiments.

[0102] Refer to the attached drawings of the specification Figure 1 , which shows a schematic flow diagram of a method for constructing a product color matching database based on data collection provided by an embodiment of the present invention.

[0103] The embodiments of the present invention provide a method for constructing a product color matching database based on data collection. This method can be implemented by a device for constructing a product color matching database based on data collection. The device for constructing a product color matching database based on data collection can be a terminal or a server. The processing flow of a method for constructing a product color matching database based on data collection can include the following steps:

[0104] S1. Multi-source data acquisition stage: Collect the surface reflection spectrum data of the target object through a distributed sensor array, and at the same time obtain the ambient light parameters and observation angle parameters, where the surface reflection spectrum data includes reflectance values at intervals of 5 nm in the range of 380 - 780 nm.

[0105] It should be noted that in the deployment of the distributed sensor array, the spatial layout of the spectral probes adopts a hexagonal grid topology, and the spacing between adjacent probes is dynamically adjusted according to the minimum feature size of the target object to ensure that the coverage density meets the requirements of the sampling theorem. The installation of the ambient light sensor needs to avoid the interference of direct light sources, and uniform sampling is achieved through a diffuse reflection cover. During the data acquisition process, a dynamic calibration mechanism is introduced, and online calibration is performed every 30 minutes using a standard whiteboard to eliminate the sensor drift error. For sudden environmental interferences (such as flashlights or local shadows), a sliding window Gaussian filtering algorithm is used to suppress the impact of transient noise on the spectral data.

[0106] In this embodiment, the reflected spectral data of the surface of the target object is collected by a distributed spectral sensor array (including at least 12 groups of high-precision spectral probes) deployed in a three-dimensional space. Each probe synchronously obtains the reflectance values at intervals of 5 nm within the wavelength range of 380 - 780 nm. At the same time, the current ambient illuminance E (unit: lx) and color temperature T (unit: K) are recorded by the ambient light sensor, and the observation angle θ (unit: degree) is obtained by a six-axis gyroscope. The data sampling frequency is set to 100 Hz to ensure data continuity in a dynamic environment.

[0107] S2. Feature dimension conversion stage: The collected original spectral data is converted into three-dimensional coordinates in the chromaticity space that conforms to human visual perception, including the brightness dimension, hue dimension, and saturation dimension, and the corresponding spectral feature vectors are generated.

[0108] It should be noted that during the chromaticity space conversion process, a 24-bit fixed-point quantization algorithm is used to control the calculation accuracy. The resolution of the brightness dimension (L*) needs to reach 0.01 cd / m 2 , and the resolution of the chromaticity coordinates (a*, b*) does not exceed 0.001. To ensure the conversion accuracy, a spectral reconstruction verification module is built into the system, and it is verified through reverse calculation whether the converted chromaticity values can restore the original spectral features, and the reconstruction error needs to be less than 2%. The GPU acceleration strategy divides the spectral bands into 16 parallel thread blocks, and each thread block processes 25 nm bandwidth data, and the real-time calculation of the tristimulus values is realized using the CUDA architecture.

[0109] In this embodiment, the original spectral data is input into the GPU-accelerated feature conversion engine to perform the following processing:

[0110] S201. Perform tristimulus value conversion using the CIE 1931 standard chromaticity observer function:

[0111]

[0112] S202. Convert to the CIELAB color space:

[0113]

[0114]

[0115] where X n , Y n , Z n are the tristimulus values under the D65 standard light source.

[0116] S3. Environment adaptation and correction stage: Calculate the environment adaptation coefficient according to the illuminance value and color temperature value in the ambient light parameters. This coefficient includes a spectral shift compensation factor and a brightness dynamic compensation factor.

[0117] It should be noted that the dynamic correction of environmental parameters uses an inertial smoothing algorithm to buffer the sudden changes in color temperature and illuminance, avoiding color distortion caused by parameter jumps. The material reflectance database contains bidirectional reflectance distribution function (BRDF) data of more than 300 materials, and the storage format covers the reflected brightness value and attenuation coefficient under the standard light source D65. For low-illuminance environments (E < 100 lx), the system automatically enables the multi-frame superposition noise reduction mode to improve the signal-to-noise ratio through time-domain integration.

[0118] In this embodiment, the environment adaptation processor executes:

[0119] Calculation of the spectral shift compensation factor δ:

[0120]

[0121] When the ambient illuminance E < 100 lx, the low-illuminance compensation mode is enabled, and at this time, the value of E is taken as max(E, 50);

[0122] Calculation of the brightness dynamic compensation factor η:

[0123]

[0124] where L D65 is obtained by querying the material reflectance database, and the value of γ is automatically selected according to the surface roughness of the material.

[0125] S4. Perceptual weight calculation stage: Based on the material type and surface texture characteristics of the target object, combined with the observation angle parameters, calculate the color perception weight matrix, which includes the visual saliency factors of each chromaticity channel.

[0126] It should be noted that for texture analysis, Daubechies 9 / 7 wavelets are used for 5-level decomposition, and the energy ratio is calculated by dividing into 6 frequency bands, with the frequency band range covering 0.5 - 32 cycles / mm. The observation angle parameter realizes motion compensation through the quaternion interpolation algorithm. When the object moving speed exceeds 0.5 m / s, the sampling frequency is increased to 200 Hz. The selection of the material basic weight vector is based on the preset material optical property mapping table. The metal material emphasizes the weight dimension of brightness, while the natural material strengthens the influence of the hue dimension.

[0127] In this embodiment, the perception weight generator executes:

[0128] Surface texture analysis: Use the VGG-19 network to extract texture frequency domain features and calculate the texture complexity index TCI:

[0129]

[0130] where f i is the central frequency of the i-th frequency band after wavelet decomposition, and e i is the energy ratio of the corresponding frequency band;

[0131] Observation angle correction: Calculate the angle correction function in real time:

[0132] f(θ) = 1 + 0.5·sin(2θ) - 0.2·cos(θ)

[0133] Generate the final weight matrix:

[0134] w final = [w L , w a , w b ·(1 + TCI)·f(θ)

[0135] S5. Multi-dimensional data fusion stage: Perform a tensor product operation on the chromaticity coordinates after environmental adaptation correction and the color perception weight matrix to generate a standardized color matching feature vector.

[0136] It should be noted that before the tensor product operation, Gram-Schmidt orthogonalization processing is performed to eliminate the correlation between chromaticity dimensions. The Tucker decomposition is used to reduce the dimension of the high-order tensor, and the core tensor corresponding to 95% energy is retained, reducing the computational complexity by two orders of magnitude. In the normalization processing stage, an adaptive standard deviation threshold is introduced, and the output range is constrained by the hyperbolic tangent function to prevent the problem of gradient saturation or disappearance.

[0137] In this embodiment, the multi-dimensional fusion core executes a tensor operation:

[0138] T ijk = ∑ m (A im·B mjk ) + C ijk ·η

[0139] where A is the environmental adaptation coefficient matrix, B is the perception weight tensor, and C is the basis tensor. The operation result is normalized by the hyperbolic tangent function:

[0140]

[0141] σ is the standard deviation of the eigenvector.

[0142] S6. Intelligent clustering analysis stage: The improved density clustering algorithm is used to classify the standardized color matching feature vectors, and a color matching database with a hierarchical structure is established. The top-level nodes store the main color features, and the child nodes store the auxiliary color features.

[0143] It should be noted that the hierarchical splitting mechanism sets a hysteresis tolerance. The generation of child nodes is triggered when HDI > 2.5 and SG < 0.8 are detected continuously for 3 times to avoid structural instability caused by instantaneous fluctuations. The initialization of the clustering center adopts the K-means++ algorithm optimized by simulated annealing, with the initial temperature set to 10 °C and converging after 50 iterations. The condition for merging child nodes is set as the sample proportion being less than 0.5% and no update occurring for 5 optimization cycles.

[0144] In this embodiment, the intelligent clustering host executes:

[0145] Core distance calculation:

[0146] core dist = 0.7·d mean + 0.3·d median

[0147] d mean is the average distance of the k = 5 nearest neighbors;

[0148] Reachable distance optimization:

[0149] reach dist = max{core dist , d(x, y)·exp(-0.05ΔE 00 )}

[0150] Hierarchical splitting determination:

[0151] When HDI > 2.5 and SG < 0.8, the generation of child nodes is triggered;

[0152] The K-means++ algorithm is used to initialize the clustering center, and the upper limit of the number of iterations is set to 200 times.

[0153] S7. Dynamic Optimization and Update Phase: According to the preference parameters and usage frequency parameters in the user feedback data, the weight distribution of each node in the color matching database is dynamically adjusted through the backpropagation algorithm.

[0154] It should be noted that a confidence evaluation mechanism is introduced for the user feedback data, and a manual review process is initiated for users with large score fluctuations (standard deviation > 2.0). During the weight adjustment process, a hue change threshold (±15°) and a saturation change threshold (±30%) are set. When the range is exceeded, the original features are retained and an exception log is generated. The Adam algorithm is used for backpropagation optimization, and the learning rate dynamic decay strategy is to decrease by 5% every 100 iterations, and the batch size is fixed at 128 samples.

[0155] In this embodiment, the dynamic optimization engine performs backpropagation:

[0156] Calculate the weight adjustment factor:

[0157]

[0158] Construct the loss function:

[0159]

[0160] The Adam optimizer is used for parameter update, the learning rate is set to 0.005, and the batch size is 128.

[0161] Refer to the attached Figure 2 illustrates the structural schematic diagram of a product color matching database construction system provided by an embodiment of the present invention.

[0162] The present invention also provides a product color matching database construction system based on data collection, which is applied to the above-mentioned product color matching database construction method based on data collection, and includes:

[0163] The spectral sensing unit in the multi-source data acquisition module 110 uses a Hamamatsu C12880MA micro-spectrometer with a wavelength resolution of 2nm; the environmental perception unit integrates an AMS AS7341 multi-channel spectral sensor and a TI TMP007 infrared temperature sensor; the spatial positioning unit uses an MPU-6050 six-axis motion processing chip with an angle measurement accuracy of ±0.01°.

[0164] The GPU acceleration unit in the feature conversion engine 120 is an NVIDIA Jetson TX2 module with a built-in 256-core Pascal architecture GPU; the chromaticity conversion chip is implemented by a Xilinx Artix-7 FPGA to realize the real-time conversion from CIE LAB to CIECAM02; the data cache is configured with 8GB LPDDR4 memory and a data transfer rate of 4266MT / s.

[0165] The spectral compensation unit 131 in the environment adaptation processor 130 is a TI TMS320C6748 DSP chip, dedicated to δ value calculation; the brightness compensation unit 132 internally stores a reflectivity database of Micron MT48LC16M16 SDRAM; the fusion output unit 133 uses an analog device AD8338 variable gain amplifier to implement analog calculation of 0.6δ + 0.4η.

[0166] The texture analysis unit in the perception weight generator 140 runs the VGG-19 network on a Cambricon MLU100 intelligent processing card; the weight calculation core is an ARM Cortex-M7 processor that performs TCI and f(θ) calculations in real time; the material feature library stores optical characteristic parameters of 200 materials and supports online updates.

[0167] The tensor operation unit in the multi-dimensional fusion core 150 is an Edge TPU coprocessor, supporting fourth-order tensor operations; the orthogonalization processor is an Intel Cyclone 10 FPGA to implement Gram-Schmidt orthogonalization; the normalization unit is an ADI ADSP-SC589 processor to execute the hyperbolic tangent function and L2 normalization.

[0168] The core distance module 161 in the intelligent clustering host 160 is a Xilinx Kintex-7 FPGA for parallel computing core dist ; the color difference calculation unit 162 is an ASIC chip to implement the CIE2000 color difference formula, with a calculation delay < 5ns; the hierarchical controller 163 is a Raspberry Pi CM4 module to run the HDI / SG monitoring algorithm.

[0169] The user behavior analyzer in the dynamic optimization engine 170 is a MongoDB database to store user preference data; the backpropagation optimizer is an NVIDIA T4 GPU to accelerate neural network training; the hybrid computing bus is a PCIe 4.0x16 interface to achieve 200Gbps data transmission between modules.

[0170] The data flow of the product color matching database construction system based on data collection provided in this embodiment is as follows:

[0171] The original spectral data is input into the feature conversion engine through a USB3.0 interface; the environment adaptation processor and the perception weight generator synchronize parameters through an SPI bus; the multi-dimensional fusion core directly accesses the GPU video memory through a DMA channel; the clustering result is transmitted to the dynamic optimization engine through a gigabit Ethernet; the optimized database is persistently stored through an NVMe SSD.

[0172] As Figure 3As shown in the figure, the distributed spectral probe array synchronously transmits the collected reflected spectral data (in the 380 - 780nm band) and the illuminance (E) and color temperature (T) parameters obtained by the environmental sensor to the feature conversion engine through the USB3.0 interface. The six - axis gyroscope continuously uploads the observation angle (θ) data at a sampling rate of 400Hz and aligns the timestamps with the spectral data through the SPI bus. After the GPU acceleration unit converts the original spectral data into CIELAB chromaticity coordinates, it is transmitted in two paths: the luminance component (L*) enters the dynamic luminance compensation unit, calculates the η value in combination with the material database, and the chromaticity components (a*, b*) are sent to the spectral offset compensation unit to calculate the δ value in real - time through the DSP chip. After the corrected parameters are weighted and fused (α = 0.6δ+0.4η), they are sent to the perception weight generator through the PCIe 3.0x8 channel. The frequency - domain features (TCI) extracted by the texture analysis unit are combined with the observation angle correction function f(θ) to generate a weight matrix, and the fourth - order tensor product operation is executed through the DMA channel directly connected to the tensor fusion coprocessor. The normalized feature vectors are transmitted to the intelligent clustering host through Gigabit Ethernet, the improved OPTICS algorithm generates a hierarchical database structure in real - time, and the optimization instructions are fed back to the dynamic optimization engine through NVLink 2.0 to complete the closed - loop learning.

[0173] The product color - matching database construction system based on data collection provided by the present invention can execute the product color - matching database construction method based on data collection described above and achieve the same or similar technical effects. To avoid repetition, the present invention will not elaborate further.

[0174] The beneficial effects brought by the technical solution provided by the embodiments of the present invention at least include:

[0175] (1) In the present invention, through the dual technical means of multi - source data fusion and environmental adaptation correction, the problem of color consistency across media and environments is innovatively solved. The environmental adaptation coefficient dynamically compensates for the influence of different lighting conditions on color perception. Combining the perception weight calculation of material characteristics and observation angles, the color - matching scheme has self - adaptability. The tensor operation in multi - dimensional data fusion constructs a non - linear mapping relationship in the high - dimensional chromaticity space, breaking through the limitations of traditional linear models. The combination of the hierarchical clustering structure and the dynamic optimization mechanism realizes the intelligent evolution ability of the color - matching database, significantly improving the accuracy and generalization of color matching, and providing a universal color - matching solution for multiple fields;

[0176] (2) In the present invention, the system adopts an architecture that deeply integrates perceptual computing and machine learning to achieve full-process intelligence from data collection to optimization decision-making. The collaborative work of the environment adaptation processor and the perceptual weight generator simulates the color adaptation mechanism of the human visual system, making the color matching results more in line with subjective perception. The improved density clustering algorithm combined with the HDI / SG index system can autonomously identify the correlation between the main color and the auxiliary color, construct a color matching database with semantic levels. The dynamic optimization engine establishes a closed-loop learning system through user feedback data, enabling the database to continuously evolve and meet the personalized needs of users, significantly reducing the dependence on manual parameter adjustment and improving the efficiency of color matching decisions;

[0177] (3) In the present invention, the heterogeneous computing architecture design at the hardware level effectively balances the processing speed and energy efficiency ratio. Through the collaborative acceleration of GPU / FPGA / ASIC, it meets the real-time requirements. The modular design supports distributed deployment and elastic expansion, and can adapt to the multi-scenario requirements from embedded devices to cloud servers. The standardized interface and cross-platform compatibility design reduce the complexity of system integration and facilitate docking with existing color management systems. The environmental robustness enhancement technology (such as dynamic calibration, anti-interference filtering) ensures stable operation under complex conditions in industrial fields, providing reliable technical support for emerging fields such as intelligent manufacturing and digital twin.

[0178] The above content is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

[0179] The following points need to be explained:

[0180] (1) The drawings of the embodiments of the present invention only relate to the structures involved in the embodiments of the present invention, and other structures can refer to the general design.

[0181] (2) For clarity, in the drawings used to describe the embodiments of the present invention, the thickness of the layers or regions is enlarged or reduced, that is, these drawings are not drawn according to the actual scale. It can be understood that when an element such as a layer, film, region or substrate is referred to as being "on" or "under" another element, the element can be "directly" on or under another element or there can be intermediate elements.

[0182] (3) Without conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other to obtain new embodiments.

[0183] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. The protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A method for constructing a product color matching database based on data collection, characterized in that, Including: S1. Multi-source data acquisition stage: The surface reflection spectrum data of the target object is collected through a distributed sensor array, and at the same time, the ambient light parameters and observation angle parameters are obtained, wherein the surface reflection spectrum data contains reflectance values at intervals of 5 nm within the range of 380 - 780 nm; S2. Feature dimension conversion stage: The collected original spectrum data is converted into three-dimensional coordinates of a chromaticity space that conforms to human visual perception, including a brightness dimension, a hue dimension, and a saturation dimension, and a corresponding spectral feature vector is generated; S3. Environment adaptation and correction stage: According to the illuminance value and color temperature value in the ambient light parameters, an environment adaptation coefficient is calculated, and this coefficient includes a spectral offset compensation factor and a brightness dynamic compensation factor; S4. Perceptual weight calculation stage: Based on the material type and surface texture characteristics of the target object, combined with the observation angle parameters, a color perception weight matrix is calculated, and this matrix includes visual saliency factors for each chromaticity channel; S5. Multidimensional data fusion stage: A tensor product operation is performed on the chromaticity coordinates after environment adaptation and correction and the color perception weight matrix to generate a standardized color matching feature vector; S6. Intelligent clustering analysis stage: An improved density clustering algorithm is used to classify the standardized color matching feature vector, and a color matching database with a hierarchical structure is established, where the top-level nodes store the main color features and the sub-nodes store the auxiliary color features; S7. Dynamic optimization and update stage: According to the preference parameters and usage frequency parameters in the user feedback data, the weight distribution of each node in the color matching database is dynamically adjusted through the backpropagation algorithm.

2. The method for constructing a product color matching database based on data collection according to claim 1, characterized in that In S3, it includes: S301. The calculation formula for the spectral offset compensation factor δ is: where, ΔT is the absolute temperature difference value between the ambient color temperature and the standard light source D65, with the unit of K, E is the current ambient illuminance value, with the unit of lx, E0 is the reference illuminance reference value of 2000 lx, CRI is the color rendering index of the ambient light source, and k1 = 0.0032, k2 = 0.15 are empirical coefficients; S302. The calculation formula for the brightness dynamic compensation factor η is: Among them, L is the measured brightness value under the current ambient illuminance, and L max is the maximum reflection brightness of the material under the standard light source, γ is the attenuation coefficient related to the material type, taking 0.12 for smooth surfaces and 0.08 for rough surfaces; S303. The environment adaptation coefficient α is finally obtained by weighted summation of δ and η: α = 0.6δ + 0.4η.

3. A method for constructing a product color matching database based on data collection according to claim 1, characterized in that, S4 specifically includes: The calculation method of the color perception weight matrix includes the following steps: S401. The frequency domain features of the object surface texture are extracted through a convolutional neural network, and the texture complexity index TCI is calculated; where freq band (i) is the center frequency value of the i-th frequency band, and energy ratio (i) is the energy percentage of the corresponding frequency band; S402. Determine the basic weight vector w according to the material type base = [w L , w a , w b , where: Metal material: w L = 0.4, w a = 0.35, w b = 0.25 Polymer material: wL = 0.3, w a = 0.4, w b = 0.3 Natural material: w L = 0.25, w a = 0.3, w b = 0.45; S403. An observation angle correction function is introduced: f(θ) = 1 + 0.5·sin(2θ) - 0.2·cos(θ) Final color perception weight matrix w final = w base ·(1 + TCI)·f(θ).

4. A method for constructing a product color matching database based on data collection according to claim 1, characterized in that S6 specifically includes: The improved density clustering algorithm adopts an improved scheme based on the OPTICS algorithm, specifically including: S601. An adaptive core distance parameter is introduced; core dist (ε) = μ·d mean +(1 - μ)·d median where d mean is the average of the k-nearest neighbor distances of the sample points, and d median is the median of the k-nearest neighbor distances of the sample points, and μ = 0.7 is the mixing coefficient; S602. A color similarity constraint term is introduced in the calculation of the reachable distance; reach dist (x,y) = max{core dist (x), d(x,y)·S(x,y)} Among them, S(x, y) = exp(-β·ΔE 00 (x, y)) is the color difference similarity function, and ΔE 00 (x, y) is the color difference value calculated by using the CIE2000 color difference formula, and β = 0.05 is the adjustment coefficient.

5. A method for constructing a product color matching database based on data collection according to claim 4, characterized in that, S6 further includes: The generation method of the middle hierarchical structure includes: S603. Calculate the main color feature extraction threshold; Among them, is the variance of the hue dimension, ρ ab is the covariance coefficient of hue and saturation, is the saturation gradient change rate; S604. Define the hue dispersion index HDI; Among them, h i is the hue value of the clustering sample, and μ h is the average hue value; S605. The calculation formula for the saturation gradient SG is: where s i is the sorted saturation sequence, and σ s is the saturation standard deviation; S606. When HDI > 2.5 and SG < 0.8, start the sub-node splitting mechanism and use the improved K-means++ algorithm to generate auxiliary color feature clusters.

6. A method for constructing a product color matching database based on data collection according to claim 1, characterized in that The dynamic optimization update in S7 specifically includes: S701. Define the weight adjustment factor ω: ω=λ·(1 - exp(-τ·f p ))+(1 - λ)·tanh(v·f u ) Among them, f p is the preference score value, ranging from 0 to 10, and f u is the usage frequency, with the unit of times per month. λ = 0.65 is the balance coefficient, and τ = 0.2, v = 0.15 are the decay coefficients; S702. Construct the backpropagation objective function: Among them, F i is the original feature vector, F' i is the optimized feature vector, ΔH i is the hue adjustment amount, ΔS i is the saturation adjustment amount, and ρ = 0.3 is the regularization coefficient; S703. Iteratively update the database node weights through the gradient descent algorithm until the objective function converges.

7. A product color matching database construction system based on data collection, characterized in that, It includes: The multi-source data acquisition module (110) is configured with a high-precision spectral sensor array and a six-degree-of-freedom attitude sensor for synchronously acquiring the object surface reflection data and spatial orientation information; The feature conversion engine (120) contains a GPU-accelerated chromaticity space converter to perform real-time conversion from CIE LAB to CIECAM02 visual perception space; The environment adaptation processor (130) is integrated with a programmable light compensation unit for calculating the environment adaptation coefficient; The perception weight generator (140) has a built-in material feature library and texture analysis algorithm to implement the calculation of the color perception weight matrix; The multi-dimensional fusion core (150) is equipped with a tensor operation coprocessor to achieve non-linear fusion of feature vectors; The intelligent clustering host (160) is equipped with a dedicated computing card with an improved density clustering algorithm; The dynamic optimization engine (170) includes a hybrid computing architecture of a user behavior analyzer and a backpropagation optimizer.

8. A method for constructing a product color matching database based on data collection according to claim 7, characterized in that The multi-dimensional fusion core specifically includes: The feature orthogonality unit (151) uses the Gram-Schmidt orthogonality method to eliminate the correlation between chromaticity dimensions; The tensor product calculation unit (152) supports fourth-order tensor operations and executes the formula: where A is the environment adaptation coefficient matrix, B is the perception weight tensor, C is the basis tensor, and η is the brightness compensation factor; The normalization output unit (153) applies a combination of the hyperbolic tangent function and L2 normalization processing to output a standardized color matching feature vector.

9. A method for constructing a product color matching database based on data collection according to claim 7, characterized in that, The environment adaptation processor specifically includes: Spectral shift compensation unit (131), calculates the δ value in real time, and the built-in DSP chip performs fast operations; A dynamic brightness compensation unit (132) is configured with a material reflectivity database and calculates an η value according to an input ratio calculation. The fusion output unit (133) realizes the hardware-level calculation of α = 0.6δ + 0.4η through a weighted summation circuit, and the output delay is less than 2 μs.

10. A method for constructing a product color matching database based on data collection according to claim 7, characterized in that The intelligent clustering host specifically includes: The core distance calculation module (161) is implemented using FPGA to achieve the parallel calculation of core dist (ε); The reachable distance optimizer (162) is integrated with an ASIC chip dedicated to chromatic aberration calculation to support the hardware acceleration of the CIE2000 chromatic aberration formula; The clustering structure generator (163) is configured with a hierarchical relationship construction algorithm to perform real-time monitoring of HDI and SG indicators and sub-node splitting control; The visualization interface (164) contains a three-dimensional chromaticity space rendering engine to support the interactive topological display of the color matching database.

Images