Simple testing method for saturated light-emitting light density of low-basis-weight thermo-sensitive paper

Abstract

The invention discloses a simple test method for saturated light-emitting light density of low-basis-weight thermo-sensitive paper, and relates to the technical field of thermo-sensitive material detection.The simple test method comprises the steps that an undeveloped surface image of to-be-tested low-basis-weight thermo-sensitive paper is obtained, and the undeveloped surface image and initial heating parameters are input into a pre-trained generative adversarial network model; generating a predicted color developing light density distribution image corresponding to the saturated color developing state and associated initial heating parameters; and carrying out weighted normalization processing on the optical density distribution data based on thin-layer structure characteristics, and outputting a global saturated optical density value and a color rendering uniformity evaluation result. According to the method, the generative adversarial network and the multispectral imaging technology are combined, so that the optical density distribution in the saturated color development state is directly predicted from the undeveloped surface image, and the test efficiency and the measurement precision are remarkably improved.

Description

Technical Field

[0001] The invention relates to the technical field of thermosensitive material detection, in particular to a simple test method for the saturated color development optical density of low-quantity thermosensitive paper. Background Art

[0002] Currently, thermal paper, as an essential medium for information recording and transmission, is widely used in healthcare, logistics, retail, and other fields. Low-weight thermal paper, due to its lightweight and environmentally friendly advantages, has gradually become a mainstream product in the market. Its saturated color optical density, a core indicator of color rendering performance, directly affects print quality and information readability. Therefore, accurate testing of the saturated color optical density of low-weight thermal paper is particularly important.

[0003] Traditional methods for measuring saturated color optical density rely primarily on physical heating color development combined with optical densitometer measurement. These methods typically rely on fixed heating parameters and single-point sampling, with color development and optical density assessment performed manually. However, these methods often suffer from significant measurement errors when faced with the thin layer structure and non-uniform coating characteristics of low-quantity thermal paper, particularly due to the dynamic coupling between heating parameters and color development. Especially given the complex color development dynamics and spatial heterogeneity, the efficiency and consistency of these traditional methods are insufficient to meet the requirements of industrial production. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a simple test method for the saturated color optical density of low-quantity thermal paper to solve the problems of large measurement errors, low efficiency and inability to adapt to industrial high-speed detection requirements of traditional test methods.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides a simple method for testing the saturated color optical density of low-weight thermal paper, comprising: obtaining an uncolored surface image of the low-weight thermal paper to be tested, and inputting the uncolored surface image and initial heating parameters into a pre-trained generative adversarial network model to generate a predicted color optical density distribution image corresponding to a saturated color state and associated initial heating parameters; inputting the predicted color optical density distribution image and the initial heating parameters into a pre-trained meta-reinforcement learning model, and constructing a dynamic reward function by real-time calculation of the optical density difference between the predicted color optical density distribution image and the physical color development process; generating a partitioned heating control strategy based on the dynamic reward function, controlling a micro-heating device to dynamically heat the thermal paper in partitions, and simultaneously acquiring multispectral time-series images of the color development process at high frequency using an optical sensor; performing a real-time difference comparison between the multispectral time-series image and the predicted color optical density distribution image; and determining that a saturated color state has been reached and terminating heating when the optical density difference between multiple consecutive frames of images is stably lower than a preset difference threshold. The optical density distribution data is then weighted normalized based on thin-layer structural characteristics to output a global saturated optical density value and a color uniformity evaluation result.

[0007] As a preferred solution of the simple test method for the saturated color optical density of the low-weight thermal paper of the present invention, the pre-trained generative adversarial network model includes the following steps: Collect historical non-colorized surface images and corresponding saturated color-developed optical density distribution images as training sets; Constructing a generator network and a discriminator network, wherein the generator network takes the uncolored surface image and the initial heating parameters as input and outputs a predicted colored optical density distribution image, and the discriminator network is used to distinguish between the real colored image and the generated image; By alternately optimizing the parameters of the generator and the discriminator, the optical density difference between the predicted color optical density distribution image and the true color image is minimized until convergence.

[0008] As a preferred solution of the simple test method for the saturated color optical density of the low-quantity thermal paper of the present invention, the pre-trained meta-reinforcement learning model includes the following steps: The collected reference color samples are used to extract a spatiotemporally continuous optical density data set. The dynamic optical density difference matrix is ​​calculated by spatially registering and pixel-level comparing the predicted color optical density distribution image with the optical density data set. Constructing a policy network including long short-term memory units, wherein the policy network takes the predicted color optical density distribution image and the dynamic optical density difference matrix as input and outputs dynamic heating parameter adjustment instructions; The policy network is iteratively trained based on the proximal policy optimization algorithm, and the dynamic heating parameter adjustment instructions are optimized by maximizing the cumulative return value of the dynamic reward function.

[0009] As a preferred solution of the simple test method for the saturated color optical density of the low-weight thermal paper of the present invention, the real-time calculation of the predicted color optical density distribution image and the optical density difference of the physical color development process to construct a dynamic reward function includes the following steps: The real-time optical density data set is converted into a real-time matrix through matrix processing, and the optical density value of each pixel in the predicted color optical density distribution image is extracted to form a prediction matrix; Calculate the time gradient change value based on the optical density change of adjacent time frames in the real-time matrix; The mean square error between the prediction matrix and the real-time matrix is ​​calculated as the basic reward value, and the reward weight is adjusted based on the time gradient change value to generate a dynamic reward function.

[0010] As a preferred solution of the simple test method for the saturated color optical density of the low-weight thermal paper described in the present invention, the partitioned dynamic heating is performed by dividing the surface of the thermal paper into equally spaced grid areas, and independently regulating each grid based on optimized parameters.

[0011] As a preferred solution of the simple test method for the saturated color optical density of the low-weight thermal paper of the present invention, the optical sensor is used to synchronously collect spectral images of visible light and near-infrared bands in a high-frequency sampling manner.

[0012] As a preferred solution of the simple test method for the saturated color optical density of the low-quantity thermal paper of the present invention, the preset difference threshold is determined by analyzing the optical density fluctuation characteristics of historical saturated color samples in combination with the statistical significance multiple.

[0013] As a preferred solution of the simple test method for the saturated color optical density of the low-quantity thermal paper described in the present invention, the weighted normalization process divides the color development area according to the coating thickness distribution of the thermal paper, and assigns differentiated weight coefficients to different areas for data fusion.

[0014] In a second aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the simple test method for the saturated color optical density of low-weight thermal paper as described in the first aspect of the present invention is implemented.

[0015] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the simple test method for the saturated color optical density of low-quantity thermal paper as described in the first aspect of the present invention is implemented.

[0016] The beneficial effects of the present invention are as follows: by combining generative adversarial networks with multispectral imaging technology, the optical density distribution of the saturated color development state can be directly predicted from the uncolored surface image, significantly improving the test efficiency and measurement accuracy. By constructing a meta-reinforcement learning model that includes spatiotemporal features and dynamically optimizing the partition heating parameters, the problem of uneven color development caused by the thin layer structure of low-quantity thermal paper is effectively solved. In addition, a weighted normalization processing method based on X-ray thickness measurement is used to accurately quantify the impact of coating thickness differences on the color development effect, making the test results more reliable. Finally, through real-time comparison and closed-loop control of high-frequency multispectral time-series images, the adaptability and stability of the test process are ensured, providing a fast and accurate quality detection method for industrial production. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 The figure is a flow chart of a simple test method for the saturated color optical density of low basis weight thermal paper.

[0019] Figure 2 Flowchart for generative adversarial network prediction.

[0020] Figure 3 This is the control flow chart for meta-reinforcement learning.

[0021] Figure 4 This is the flow chart of multispectral time-series image analysis. DETAILED DESCRIPTION

[0022] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0023] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0025] Reference Figures 1 to 4 , is an embodiment of the present invention, which provides a simple test method for the saturated color optical density of low-weight thermal paper, comprising the following steps: S1: Obtain an uncolored surface image of the low-quantity thermal paper to be tested, and input the uncolored surface image and initial heating parameters into a pre-trained generative adversarial network model to generate a predicted color development optical density distribution image corresponding to the saturated color development state and the associated initial heating parameters.

[0026] Specifically, the following steps are included: S1.1: Use a multispectral imaging device (such as SpecimIQ) to scan the surface of the low-weight thermal paper to be tested, and collect historical uncolored surface images and corresponding saturated color optical density distribution images as a training set; Among them, the color optical density distribution image refers to the standard optical density distribution image of thermal paper in the fully colored state obtained through experimental measurement. It is used as a training supervision label in pair with the uncolored surface image to guide the generator in the generative adversarial network to learn the mapping relationship from the input uncolored image to the saturated colored state.

[0027] Furthermore, the method for obtaining the saturated color optical density distribution includes: placing the thermal paper sample to be tested under standard color development conditions for sufficient heating treatment, using a multispectral imaging device to collect the surface optical density distribution data after the color development is completed, and forming a standardized saturated color optical density distribution image after denoising and normalization. The saturated color optical density distribution image has the same spatial resolution and optical density quantification standard as the predicted color optical density distribution image output by the generator, but the former is the actual color development result obtained by experimental measurement, and the latter is the predicted result generated by the model based on the input parameters.

[0028] Furthermore, the collected historical non-colored surface images are preprocessed.

[0029] Specifically, high-frequency noise is removed through Gaussian filtering, and the multispectral band image is converted to a grayscale image and resized to maintain the spatial resolution of the original surface texture. A multispectral band image is a collection of images collected at the same spatial location using a multispectral imaging device and containing spectral information from multiple narrow bands (usually 10 to 100 bands).

[0030] S1.2: Construct a generator network and a discriminator network. The generator network takes the uncolored surface image and the initial heating parameters as input and outputs a predicted colored optical density distribution image. The discriminator network distinguishes between the real colored image and the generated image. Furthermore, the steps for constructing and training the generative adversarial network model structure are as follows: For the generator network construction and input processing, specifically: the pre-processed uncolored grayscale image (single channel) and the heat map (dual channel) generated by the initial heating parameters are spliced ​​to form three-channel input data. Among them, the initial heating temperature includes the initial heating temperature and initial heating time .

[0031] The U-Net architecture consists of a 4-layer encoder and a 4-layer decoder, using skip connections to preserve spatial details. Encoder (downsampling): Layer 1: 3×3 convolution kernel, stride 2, 64 output channels, activation function LeakyReLU (slope 0.2); Layers 2-4: The number of channels is doubled layer by layer (128 → 256 → 512), while the convolution kernel and stride remain unchanged. Decoder (upsampling): Layers 1-3: 3×3 deconvolution kernel, stride 2, 256 → 128 → 64 output channels, activation function ReLU; Layer 4: 3×3 deconvolution kernel, stride 2, 1 output channel (single-channel optical density image), activation function Sigmoid.

[0032] The output feature map of each layer of the encoder is channel-wise spliced ​​with the input of the corresponding layer of the decoder to achieve detail fusion, and finally output a single-channel predicted color optical density distribution image.

[0033] For discriminator network construction and input processing, the discriminator receives either the actual color image or the predicted color optical density distribution image output by the generator. A four-layer convolutional network is used: Layer 1: 4×4 convolution kernel, stride 2, 64 output channels, activation function LeakyReLU (slope 0.2); Layers 2-4: The number of channels is doubled layer by layer (128 → 256 → 512), while the convolution kernel and stride are kept constant; the final layer: global average pooling followed by a fully connected layer, outputting a single-node probability value (sigmoid activation).

[0034] The output value represents the probability that the input image is a true color image and is used to calculate the adversarial loss.

[0035] S1.3: Furthermore, by alternately optimizing the parameters of the generator and the discriminator, the optical density difference between the predicted color optical density distribution image and the true color image is minimized until convergence.

[0036] Among them, the optical density difference value refers to the difference in optical density values ​​of corresponding pixels in the predicted color optical density distribution image and the actual color image, and is a quantitative evaluation result generated based on the mean square error algorithm, which is used to guide the parameter optimization process of the generator network.

[0037] Specifically, the first stage (generator freezing): update the discriminator parameters to minimize the sum of the real image discrimination loss and the generated image discrimination loss; the second stage (discriminator freezing): update the generator parameters to jointly optimize the adversarial loss (deceiving the discriminator) and the mean square error of the optical density ( ).

[0038] Optical density mean square error calculation formula: ; in, Indicates the total number of pixels, represents the pixel index, For the The predicted optical density value of pixels, For the The true optical density value of each pixel, represents the predicted value, Indicates the measured value; Training terminates when the MSE fluctuation is less than 0.001 for 10 consecutive epochs. An epoch represents the iteration cycle during which the Generative Adversarial Network (GAN) fully traverses the training set.

[0039] Specifically, each epoch includes forward propagation and backpropagation: forward propagation refers to the input of all training samples (uncolored images + heating parameters) into the generator and discriminator; backpropagation refers to updating the model parameters according to the loss function (MSE + adversarial loss).

[0040] The preprocessed uncolored surface image to be tested and the initial heating parameters are input into the generator network, and the predicted colored optical density distribution image is output.

[0041] Furthermore, the initial heating parameters are optimized: Specifically, the 512-dimensional feature vector of the last layer of the generator encoder is extracted, and the heating parameter adjustment amount is output through the fully connected layer. The calculation formula is as follows: ; ; in, represents the optimized heating temperature, Indicates the temperature adjustment amount, Indicates the optimized heating time, Indicates the time adjustment amount, It is only used to identify the optimized parameters and does not participate in mathematical operations. Its technical essence is reflected in the dynamic adjustment logic. It should be noted that the optimized heating temperature is generated by linearly superimposing the initial heating temperature and the temperature adjustment. The initial heating temperature is set to 100°C by default and ranges from 80 to 120°C. The temperature adjustment is limited to a range of [-20, 20]°C, and an empirical coefficient is used to control the adjustment amplitude to avoid overshoot. Similarly, the optimized heating time is generated by proportionally superimposing the initial heating time and the time adjustment. The initial heating time is set to 1.5 seconds by default and ranges from 0.5 to 3.0 seconds. The time adjustment is limited to a range of [-0.5, 0.5] seconds, and a smoothing coefficient is used to ensure the continuity of parameter changes. Both the temperature and time adjustments are dynamically generated by a meta-reinforcement learning model based on real-time feedback of optical density differences and are subject to zone heating safety constraints, ensuring dynamic optimization of heating parameters while balancing device stability and prediction accuracy.

[0042] We randomly select ten 50×50 pixel regions of the predicted color optical density distribution image and calculate the regional mean square error (MSE). If the MSE exceeds 0.05 OD², we trigger generator fine-tuning: we freeze the discriminator parameters and update the generator at a learning rate of 0.00002 until MSE ≤ 0.05 or 100 iterations are reached. OD² represents the square of the optical density unit and is used to quantify the mean square error (MSE) between the predicted and true color images.

[0043] S2: The predicted color optical density distribution image and initial heating parameters are input into the pre-trained meta-reinforcement learning model, and a dynamic reward function is constructed by real-time calculation of the optical density difference between the predicted color optical density distribution image and the physical color development process.

[0044] Specifically, the following steps are included: S2.1: Extract a spatiotemporally continuous optical density dataset from a benchmark colorimetric sample acquired by a multispectral imaging device. Calculate a dynamic optical density difference matrix by spatially registering and pixel-wise comparing the predicted colorimetric optical density distribution image with the real-time optical density dataset. The initial heating parameters are encoded into a two-dimensional heat map (such as a temperature heat map and a time heat map) that matches the spatial resolution of the predicted color optical density distribution image. The predicted color optical density distribution image and the two-dimensional heat map are channel-wise spliced ​​to output three-channel data as the input data of the pre-trained meta-reinforcement learning model.

[0045] The optical density dataset refers to a structured data set collected by a multispectral imaging device, encompassing spatial distribution, time series, and spectral dimensions. The three-channel data includes a predicted color optical density distribution image, a temperature heat map, and a time heat map.

[0046] S2.2: Spatially align the three-channel data after channel splicing with the dynamic optical density difference matrix. After unifying the feature dimensions through a 1×1 convolution kernel, splice it along the channel direction into four-channel fused data (the original three channels + the difference matrix channel) as the complete input of the policy network.

[0047] Constructing a policy network including long short-term memory units, wherein the policy network takes the predicted color optical density distribution image and the dynamic optical density difference matrix as input and outputs dynamic heating parameter adjustment instructions; Specifically, the policy network adopts an encoder-decoder architecture. The encoder part extracts spatial features through three convolutional layers. The size of the convolution kernel in each layer is fixed at 3×3 and the number of channels doubles layer by layer. The long short-term memory unit maintains the number of hidden units that matches the number of output channels of the convolution layer when processing temporal features. The decoder outputs dynamic heating parameter adjustment instructions through the fully connected layer; the training process adopts the proximal policy optimization algorithm, the learning rate is set to the standard recommended value, the confidence interval is set using conventional reinforcement learning, the advantage function is calculated by sampling the complete trajectory, and the dynamic reward function is composed of a linear combination of the mean square value of the prediction error and the change in the time gradient. The training termination condition is determined based on the fluctuation amplitude of the cumulative return of consecutive evaluation rounds.

[0048] The policy network is iteratively trained based on the proximal policy optimization algorithm, and the dynamic heating parameter adjustment instructions are optimized by maximizing the cumulative return value of the dynamic reward function.

[0049] Specifically, the output constraint of the dynamic heating parameter adjustment instruction is implemented through the hyperbolic tangent activation function, and its output range corresponds to the temperature adjustment range and time adjustment range allowed by the device; during the training phase, the real-time matrix is ​​subjected to fixed-ratio spatiotemporal cropping, and random noise conforming to the normal distribution is injected. Gradient norm truncation is performed during policy updates, and the experience replay buffer automatically filters and stores samples according to the degree of deviation between the reward value and the mean. All operating parameters are derived from the standard reinforcement learning implementation scheme and the physical limitations of the thermal paper color development device.

[0050] S2.3: Converting the optical density data set into a real-time matrix through matrix processing, and extracting the optical density value of each pixel in the predicted color optical density distribution image to form a prediction matrix; Among them, matrix processing includes band registration and time alignment of multispectral time series images, and then stacking them according to the space-time dimension to form a three-dimensional tensor.

[0051] Calculate the time gradient change value based on the optical density change of adjacent time frames in the real-time matrix; Specifically, based on the time-series optical density data of the color development process collected by the multispectral imaging device and after generating a real-time matrix, the optical density data of the current frame and the previous frame are extracted from the real-time matrix as input, and an element-by-element subtraction operation is performed on the corresponding pixel points of adjacent time frames to obtain an optical density change matrix, and the optical density change matrix is ​​divided by the time interval to calculate the time gradient change value per unit time.

[0052] It should be noted that dividing the optical density variation matrix by the time interval means dividing each element in the real-time matrix (ie, the optical density variation of each pixel) by the same time interval value.

[0053] The mean square error between the prediction matrix and the real-time matrix is ​​calculated as the basic reward value, and the reward weight is adjusted based on the time gradient change value to generate a dynamic reward function.

[0054] Specifically, the pixel-by-pixel square difference summation method is used to calculate the mean square error between the prediction matrix and the real-time matrix as the basic reward value. At the same time, based on the pre-established weight mapping relationship, the time gradient change value is converted into a regional adaptive weight coefficient. The basic reward value and the time gradient weight coefficient are fused through a linear weighted method. The weight coefficient is automatically reduced in the high gradient change area to reduce the impact of transient noise, and the weight coefficient is increased in the low gradient change area to enhance the steady-state accuracy evaluation, thereby generating a dynamic reward function.

[0055] It should be noted that the pre-established weight mapping relationship is a quantitative conversion relationship constructed based on the color development kinetics characteristics of thermosensitive materials. This quantitative conversion relationship is established through a standardized data acquisition and analysis process under standard test conditions. Specifically, within the time gradient change range corresponding to the active stage of the color development reaction, an adjustment coefficient in the range of 0.5-0.8 is used to reduce the impact of instantaneous fluctuations; within the time gradient change range corresponding to the stable stage of color development, an adjustment coefficient in the range of 1.2-1.5 is used to enhance the steady-state accuracy assessment.

[0056] Furthermore, the quantitative conversion relationship is parameter-fitted using a benchmark dataset, and the input-output mapping is achieved using a piecewise cubic spline interpolation method. The input is the normalized time gradient change value, and the output is the corresponding dynamic adjustment coefficient. It is configured as a preset parameter in the reward calculation process during initialization.

[0057] Preferably, the weight mapping relationship is based on the time gradient-weight conversion rule established by the color dynamic characteristics, which enables the dynamic reward function to achieve two-stage adaptive adjustment: automatically reducing the transient interference weight during the active reaction period and improving the precision control weight during the stable period.

[0058] It should also be noted that dynamic heating parameter adjustment commands are converted into zoned temperature control signals via the micro-heating device's actuator interface, driving the thermal paper's color-developing areas to adjust temperature and time according to the command parameters. The multispectral imaging device collects the adjusted color development state in real time, using the resulting real-time optical density matrix along with the current heating parameter adjustment command as feedback input, forming a closed-loop control flow of "command execution → color development response → data acquisition → reward calculation." The heating parameter adjustment command directly affects the optical density trend of the real-time matrix, and its effect is quantitatively evaluated using the time gradient change value and the predicted-actual difference matrix.

[0059] S3: Generate a partitioned heating control strategy based on the dynamic reward function, control the micro-heating device to dynamically heat the thermal paper in partitions, and simultaneously use the optical sensor to collect multi-spectral time-series images of the color development process at high frequency.

[0060] Specifically, the following steps are included: S3.1: Divide the thermal paper surface into equally spaced grid regions, with the grid divisions being spatially aligned with the grid divisions of the undeveloped surface image.

[0061] Based on the heating parameter adjustment instructions generated by the dynamic reward function, the heating power and duration of each grid area are independently controlled.

[0062] Specifically, the differentiated heating parameter adjustments for each grid area are calculated based on the dynamic reward function. These differentiated heating parameter adjustments reflect the degree of deviation between the color rendering status of different areas and the predicted target. The adjustments are superimposed on the basic heating parameters to generate independent temperature control values ​​and heating time values ​​for each grid. The driving circuit of the micro-heating device generates a specific current intensity and power-on duration for the heating unit corresponding to each grid based on these independent parameter values ​​(independent temperature control values ​​and heating time values).

[0063] It should be noted that the current intensity determines the instantaneous heating power, and the power-on time controls the heating duration; the heating unit array adopts an independent addressing design to ensure that each unit can accurately execute the power and duration combination assigned to it; the optical sensor collects the color status of each grid in real time, and the feedback data is used to verify the control effect and trigger the next round of parameter adjustment to form a closed-loop control.

[0064] S3.2: The heating unit array of the micro-heating device corresponds one-to-one to the grid area on the surface of the thermal paper, and each heating unit receives the zoned heating control strategy of the corresponding grid.

[0065] Specifically, the zoned heating control strategy obtains dynamic heating parameter adjustment instructions for each grid, loads the initial heating parameters as a baseline, and generates optimized real-time heating parameters through linear superposition, limiting the calculated results to the device's allowable range. Based on the characteristics of the heating unit, the parameters are converted into drive signals, independent control instructions are sent to each grid heating unit, and multispectral images collected by optical sensors are received for the next round of adjustments.

[0066] Partitioned dynamic heating is controlled by temperature-time dual parameters: the heating temperature is dynamically adjusted based on the optimized heating temperature, and the heating time is dynamically adjusted based on the optimized heating time. Both are constrained by the dynamic heating parameter adjustment instructions output by the meta-reinforcement learning model.

[0067] S3.3: The optical sensor synchronously collects the visible light band spectral image and the near-infrared band spectral image during the color development process using a high-frequency sampling method to generate a multi-spectral time series image with time series alignment.

[0068] The acquisition frequency of the multispectral time series images is synchronized with the heating parameter update frequency to ensure that at least one complete multispectral data acquisition is completed in each heating control cycle.

[0069] S4: Perform real-time difference comparison between the multispectral time series image and the predicted color optical density distribution image. When the optical density difference value of multiple consecutive frames of images is stably lower than the preset difference threshold, it is determined that the saturated color state has been reached and heating is terminated, and the optical density distribution data is extracted.

[0070] Specifically, the following steps are included: S4.1: Spatially register the multispectral time-series image with the predicted color optical density distribution image, and calculate the optical density difference value pixel by pixel.

[0071] Specifically, the multispectral time series images are geometrically corrected to eliminate deformation errors in the acquisition process. The feature point matching algorithm is used to align the coordinate system of the corrected multispectral time series images with the predicted color optical density distribution image. The bilinear interpolation method is used to uniformly adjust the resolution of the aligned images to the same pixel size. A pixel coordinate mapping relationship table is established to record the multispectral image sampling position corresponding to each predicted image pixel point. The radiation brightness value of each band of the multispectral time series image at the corresponding position is extracted according to the mapping relationship table. The radiation brightness value is converted into a standard optical density value and subtracted from the predicted optical density value of the corresponding pixel in the predicted image. The absolute value is taken to obtain the optical density difference value of each pixel point.

[0072] The optical density difference value is defined as the optical density value of each pixel in the predicted color optical density distribution image minus the absolute value of the optical density value of the current frame in the multispectral time series image at the corresponding position S4.2: When the optical density difference value of the continuous multiple frames of images is stably lower than the preset difference threshold, it is determined that the saturated color development state is reached and the heating is terminated.

[0073] It should be noted that the preset difference threshold is determined by analyzing the optical density fluctuation characteristics of historical saturated hair color samples and combining the statistical significance multiple.

[0074] Specifically, multiple batches of optical density distribution data of historical saturated hair color samples are collected, and the collected data are denoised and normalized to obtain a standardized optical density value matrix. The optical density fluctuation range of each sample in the saturated state is calculated as the baseline fluctuation feature. The standard deviation of the baseline fluctuation feature is calculated using a statistical analysis method and multiplied by a preset significance multiple as the initial threshold. The applicability of the initial threshold is tested under different color development conditions through cross-validation, and finally the initial threshold with the lowest misjudgment rate is selected as the officially adopted preset difference threshold.

[0075] S4.3: After heating is terminated, the last frame of the current multispectral time series image is extracted as the final optical density distribution data.

[0076] The optical density distribution data contains a matrix of standardized optical density values ​​whose spatial and spectral resolutions are consistent with the predicted color optical density distribution image. S5: Perform weighted normalization processing on the optical density distribution data based on the thin layer structure characteristics, and output the global saturated optical density value and color uniformity evaluation results.

[0077] Specifically, the following steps are included: S5.1: Establish a thickness distribution model based on the X-ray fluorescence thickness measurement data of thermal paper, and divide the inspection area into three thickness ranges: high, medium, and low. The interval division standard is consistent with the coating tolerance range in the thermal paper production process.

[0078] Specifically, an X-ray fluorescence thickness gauge is used to scan the surface of the thermal paper to be tested to obtain an original thickness data dot matrix, and the original data dot matrix is ​​Gaussian filtered to eliminate measurement noise and generate a smooth thickness distribution surface. The division standard is determined according to the coating thickness grading threshold specified in the thermal paper production process standard. The area above the upper limit coating thickness grading threshold in the smooth thickness distribution surface is marked as a high thickness interval, the area between the upper and lower limit coating thickness grading thresholds is marked as a medium thickness interval, and the area below the lower limit coating thickness grading threshold is marked as a low thickness interval. A digital thickness distribution model containing the spatial coordinates of the high / medium / low thickness intervals is output for subsequent weighted normalization processing.

[0079] It should be noted that the coating thickness grading threshold refers to the thickness range boundary value pre-set according to the thermal paper production process standard, which is used to quantify the continuously distributed coating thickness into three discrete grade intervals: high, medium and low.

[0080] Furthermore, the coating thickness classification threshold includes two key parameters: the upper coating thickness classification threshold (the critical value for distinguishing high / medium thickness intervals) and the lower coating thickness classification threshold (the critical value for distinguishing medium / low thickness intervals). The specific value is determined by the characteristics of the thermal paper substrate, the color development layer formula and the coating process parameters, and is verified and solidified by measuring the standard sample during the equipment calibration phase. In practical applications, the coating thickness classification threshold needs to match the spatial resolution of the multispectral imaging device to ensure that the division result of each thickness interval is aligned with the physical boundary of the color development area. The weighted normalization process divides the color development area according to the coating thickness distribution of thermal paper, and assigns differentiated weight coefficients to different areas for data fusion.

[0081] S5.2: Assign normalized weight coefficients to each of the high, medium, and low thickness ranges (e.g., 1.2 for the high thickness range, 1.0 for the medium thickness range, and 0.8 for the low thickness range. These normalized weight coefficients are empirically derived by least squares fitting based on the correlation between coating thickness and color sensitivity for 100 sets of standard samples). Multiply the dynamic optical density difference matrix by the weight coefficient matrix element-by-element. Perform a global summation of the weighted optical density values ​​and divide them by the total number of valid pixels to obtain the normalized global saturation optical density value.

[0082] It should be noted that the weight coefficient matrix refers to the registration matrix of the same dimension as the optical density difference matrix, which is filled with normalized weight coefficients based on the spatial coordinate mapping relationship of the thermal paper thickness distribution model. This is used to implement the data fusion operation of weighted thickness features. The spatial coordinate mapping relationship of the thermal paper thickness distribution model is the geometric transformation correspondence established between the thickness data point array coordinates obtained by the X-ray fluorescence thickness gauge and the pixel coordinates of the multispectral imaging system. This geometric transformation correspondence is achieved through the registration of the fiducial marker points on the calibration plate, ensuring that each thickness measurement point is precisely aligned with the image pixel position.

[0083] S5.3: Calculate the coefficient of variation (standard deviation to mean) of the optical density values ​​for the high, medium, and low thickness ranges, and select the maximum value as the core evaluation metric. Determine the final color uniformity rating based on pre-set thresholds (e.g., 0-0.05 for Grade A, 0.05-0.1 for Grade B, and so on).

[0084] The rating results are integrated with the global saturation optical density value, generating a structured output according to a pre-set report template. This report strictly includes two independently verified core indicators: the global saturation optical density value (normalized to three decimal places) and the color uniformity grade (based on a five-level evaluation system). The data field arrangement, unit notation, and file format are fully compatible with the storage specifications of historical testing databases, ensuring data traceability and horizontal comparability. The five-level evaluation system refers to the quality evaluation standard that divides the color uniformity evaluation results into five preset grades: A / B / C / D / E.

[0085] This embodiment also provides a computer device suitable for a simple test method for the saturated color optical density of low-weight thermal paper, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the simple test method for the saturated color optical density of low-weight thermal paper proposed in the above embodiment.

[0086] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0087] This embodiment also provides a storage medium having a computer program stored thereon. When the program is executed by a processor, the processor implements the simple test method for the saturated color optical density of low-weight thermal paper as proposed in the above embodiment. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0088] In summary, the present invention combines generative adversarial networks with multispectral imaging technology to achieve direct prediction of the optical density distribution of the saturated color development state from the uncolored surface image, significantly improving the test efficiency and measurement accuracy. By constructing a meta-reinforcement learning model that includes spatiotemporal features and dynamically optimizing the partition heating parameters, the problem of uneven color development caused by the thin layer structure of low-quantity thermal paper is effectively solved. In addition, a weighted normalization processing method based on X-ray thickness measurement is used to accurately quantify the impact of coating thickness differences on the color development effect, making the test results more reliable. Finally, through real-time comparison and closed-loop control of high-frequency multispectral time-series images, the adaptability and stability of the test process are ensured, providing a fast and accurate quality detection method for industrial production.

[0089] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A simple test method for the saturated color optical density of low-weight thermal paper, characterized by: include, Obtaining an uncolored surface image of the low-weight thermal paper to be tested, and inputting the uncolored surface image and initial heating parameters into a pre-trained generative adversarial network model to generate a predicted color development optical density distribution image corresponding to a saturated color development state and associated initial heating parameters; The predicted color optical density distribution image and initial heating parameters are input into the pre-trained meta-reinforcement learning model, and a dynamic reward function is constructed by calculating the optical density difference between the predicted color optical density distribution image and the physical color development process in real time. Based on the dynamic reward function, a zone heating control strategy is generated to control the micro-heating device to dynamically heat the thermal paper in zones. At the same time, the optical sensor is used to collect multi-spectral time-series images of the color development process at high frequency. The multispectral time series image is compared with the predicted color optical density distribution image in real time. When the optical density difference value of the continuous multi-frame image is stably lower than the preset difference threshold, it is determined that the saturated color state has been reached and the heating is terminated, and the optical density distribution data is extracted; The optical density distribution data is weighted and normalized based on the thin layer structure characteristics, and the global saturated optical density value and color uniformity evaluation results are output.

2. The simple test method for the saturated color optical density of low-weight thermal paper according to claim 1, wherein: The pre-trained generative adversarial network model includes the following steps: Collect historical non-colorized surface images and corresponding saturated color-developed optical density distribution images as training sets; Constructing a generator network and a discriminator network, wherein the generator network takes the uncolored surface image and the initial heating parameters as input and outputs a predicted colored optical density distribution image, and the discriminator network is used to distinguish between the real colored image and the generated image; By alternately optimizing the parameters of the generator and the discriminator, the optical density difference between the predicted color optical density distribution image and the true color image is minimized until convergence.

3. The simple test method for the saturated color optical density of low-weight thermal paper according to claim 2, wherein: The pre-trained meta-reinforcement learning model includes the following steps: The collected reference color samples are used to extract a spatiotemporally continuous optical density data set. The dynamic optical density difference matrix is ​​calculated by spatially registering and pixel-level comparing the predicted color optical density distribution image with the optical density data set. Constructing a policy network including long short-term memory units, wherein the policy network takes the predicted color optical density distribution image and the dynamic optical density difference matrix as input and outputs dynamic heating parameter adjustment instructions; The policy network is iteratively trained based on the proximal policy optimization algorithm, and the dynamic heating parameter adjustment instructions are optimized by maximizing the cumulative return value of the dynamic reward function.

4. The simple test method for the saturated color optical density of low-weight thermal paper according to claim 3, wherein: The real-time calculation and prediction of the optical density difference between the color development optical density distribution image and the physical color development process to construct a dynamic reward function includes the following steps: The real-time optical density data set is converted into a real-time matrix through matrix processing, and the optical density value of each pixel in the predicted color optical density distribution image is extracted to form a prediction matrix; Calculate the time gradient change value based on the optical density change of adjacent time frames in the real-time matrix; The mean square error between the prediction matrix and the real-time matrix is ​​calculated as the basic reward value, and the reward weight is adjusted based on the time gradient change value to generate a dynamic reward function.

5. The simple test method for saturated color optical density of low-weight thermal paper according to claim 1, wherein: The zoned dynamic heating is achieved by dividing the surface of the thermal paper into equally spaced grid areas and independently regulating each grid area based on optimized parameters.

6. The simple test method for the saturated color optical density of low-weight thermal paper according to claim 1, wherein: The optical sensor is a device that synchronously collects spectral images of visible light and near-infrared bands in a high-frequency sampling manner.

7. The simple test method for saturated color optical density of low-weight thermal paper according to claim 1, wherein: The preset difference threshold is determined by analyzing the optical density fluctuation characteristics of historical saturated hair color samples and combining the statistical significance multiple.

8. The simple test method for saturated color optical density of low-weight thermal paper according to claim 1, wherein: The weighted normalization process divides the color development area according to the coating thickness distribution of the thermal paper, and assigns differentiated weight coefficients to different areas for data fusion.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the simple test method for the saturated color optical density of the low-weight thermal paper according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the simple test method for the saturated color optical density of low-weight thermal paper according to any one of claims 1 to 8 are implemented.

Images