Spectroscopic detection device
The spectral reconstruction model built by the autoencoder and detection network solves the problems of high price, slow speed and large recognition error of traditional spectral detection equipment, and realizes efficient and accurate spectral detection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANGHAI PUBLISHING & PRINTING COLLEGE
- Filing Date
- 2021-07-07
- Publication Date
- 2026-06-26
AI Technical Summary
In existing technologies, spectral detection equipment is expensive, has high maintenance costs, and is slow to detect. The spectral reconstruction model is also large in structure, which can easily lead to identification errors.
A spectral reconstruction model is constructed using an autoencoder and a detection network. The detection network is trained through image acquisition, color value extraction, clustering, and sorting to achieve efficient reconstruction of spectral reflectance.
It improves the speed and accuracy of spectral detection, reduces the complexity of model structure, optimizes the prediction ability of untrained colors, and reduces detection costs.
Smart Images
Figure CN115601287B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of color detection technology, and more specifically to a spectral detection device. Background Technology
[0002] In color-related applications such as printing production, printing quality control, and artwork reproduction, it is necessary to obtain quality control parameters such as chromaticity, brightness, saturation, solid density, dot area ratio, and dot gain of color blocks based on their spectral reflectance. These quality control parameters are then used to control product quality during the production process.
[0003] Traditionally, the spectral reflectance of color patches is detected using specialized measuring equipment, such as a spectrometer, and then quality control parameters are calculated. However, such measuring equipment is expensive and has high maintenance costs, the testing process is slow, and the test results have significant identification errors.
[0004] To address the aforementioned issues, existing technology (CN110135523A) can detect the spectrum using a camera. Specifically, it uses a camera to capture an image to be tested, which includes multiple standard color patches. The RGB values of the standard color patches are then extracted, and the corresponding spectral reflectance is obtained by reconstructing the RGB values using a pre-built and trained spectral reconstruction model.
[0005] However, this method still has problems. Specifically, since spectral reflectance includes spectral values across multiple different bands (generally 31 bands), when reconstructing the spectrum using the above method, the model needs to reconstruct all 31 spectral values based on the RGB values. This can lead to two problems: firstly, the spectral reconstruction model becomes very large, reducing the reconstruction speed and thus affecting the detection speed; secondly, the spectral reconstruction model still has a certain possibility of producing recognition errors during the reconstruction of spectral reflectance. Summary of the Invention
[0006] To address the aforementioned problems, this invention provides a spectral detection device with high detection speed and high recognition accuracy during spectral detection. The technical solution adopted by this invention is as follows:
[0007] This invention provides a spectral detection device for detecting printed colors. It comprises: an image acquisition unit for acquiring a target image obtained by image acquisition of a target object, the target image containing at least one color patch to be tested; a color value extraction unit for extracting the color value of the color patch to be tested as the color value to be tested; a spectral reconstruction unit having a pre-trained spectral reconstruction model for reconstructing the corresponding spectral reflectance based on the color value to be tested; and an output unit for outputting the spectral reflectance. The spectral reflectance includes multiple bands and the spectral value of each band. The spectral reconstruction model has an autoencoder and a detection network. The autoencoder has a code for encoding the spectral reflectance of multiple bands into a predetermined number of classes of encoded data and a corresponding decoder for decoding the encoded data into spectral reflectance. The detection network is pre-trained through the following steps: Step S1, acquiring a standard image obtained by the image acquisition unit scanning a standard printing template, the standard image containing multiple standard color patches; Step S2, acquiring the color value of the standard color patch as the standard color value and the corresponding spectral reflectance as the standard spectral reflectance; Step S3, clustering the standard color values and dividing the multiple standard color values into groups. The system is divided into several color clusters; Step S4: The standard color values in each color cluster are sorted according to their size to form several ordered color clusters; Step S5: The standard spectral reflectance corresponding to each standard color value is sequentially input into the autoencoder to obtain the corresponding encoded data as standard encoded data; Step S6: The detection network is trained sequentially based on the standard color value and the corresponding standard encoded data according to the order of the standard color values in each ordered color cluster until the predetermined conditions are met and a trained detection network is obtained. The trained detection network can obtain the corresponding predicted encoded data based on the color value to be tested. The predicted encoded data can be decoded by the decoder into the spectral reflectance corresponding to the color value to be tested.
[0008] The spectral detection device provided by the present invention may also have the following technical features, wherein the spectral reflectance has 31 bands and a predetermined number of categories of 3 to 12.
[0009] The spectral detection device provided by the present invention may also have the following technical feature, wherein the predetermined number of categories is 3 or 8.
[0010] According to the spectral detection device provided by the present invention, it may also have the following technical features, wherein the predetermined condition is to determine whether the detection effect of the detection network meets a predetermined threshold, and to test the trained detection network using a predetermined test dataset; to determine whether the test result meets the predetermined threshold, and if the determination result is yes, the detection network is successfully trained; if the determination result is no, step S6 is repeated to train the detection network again.
[0011] According to the spectral detection device provided by the present invention, the color patch to be tested is a rectangular color patch with a width between 0.05 mm and 2 mm.
[0012] According to the spectral detection device provided by the present invention, it may also have the following technical features, wherein the color value is an RGB value, and step S4 includes the following sub-steps: step S4-1, sorting each color cluster sequentially according to the size of the RGB values to form an RGB value sequence; step S4-2, extracting a predetermined number of RGB values from the RGB value sequence at predetermined intervals, and forming ordered color clusters from the extracted RGB values.
[0013] This invention also provides a spectral detection method for spectral detection of the color of a target object, characterized by comprising: an image acquisition step, acquiring a target image obtained by image acquisition of the target object, the target image having at least one color patch to be tested; a color value extraction step, extracting the color value of the color patch to be tested as the color value to be tested; a spectral reconstruction step, inputting the color value to be tested into a pre-trained spectral reconstruction model, and reconstructing the spectral reflectance corresponding to the color value to be tested by the spectral reconstruction model; and an output step, outputting the spectral reflectance; wherein the spectral reflectance includes multiple bands and the spectral value of each band, the spectral reconstruction model has an autoencoder and a detection network, the autoencoder has a decoder for encoding the spectral reflectance with multiple bands into a predetermined number of encoded data and a corresponding decoder for decoding the encoded data into spectral reflectance, and the detection network is pre-trained through the following steps: Step S1, the image acquisition unit scans a standard printing template to obtain... The standard image obtained contains multiple standard color patches; Step S2: Obtain the color value of the standard color patch as the standard color value and the corresponding spectral reflectance as the standard spectral reflectance; Step S3: Cluster the standard color values to divide the multiple standard color values into several color clusters; Step S4: Sort the standard color values in each color cluster according to their size to form several ordered color clusters; Step S5: Input the standard spectral reflectance corresponding to each standard color value into the encoder to obtain the corresponding encoded data as standard encoded data; Step S6: Train the initial detection network according to the order of the standard color values in each ordered color cluster, based on the standard color value and the corresponding standard encoded data, until the predetermined conditions are met and a trained detection network is obtained. The trained detection network can obtain the corresponding predicted encoded data based on the color value to be tested, and the predicted encoded data can be decoded by the decoder into the spectral reflectance corresponding to the color value to be tested.
[0014] Invention Function and Effect
[0015] According to the spectral detection apparatus of the present invention, the color value of the color patch to be tested is extracted from the target image scanned by the image acquisition unit by the color value extraction unit, and spectral reconstruction is performed by the spectral reconstruction unit. The spectral reconstruction unit has a pre-trained spectral reconstruction model, which includes a detection network and an autoencoder. The autoencoder has an encoder for encoding spectral reflectance of multiple bands into a predetermined number of coded data classes and a corresponding decoder for decoding the coded data into spectral reflectance. Therefore, the spectral reconstruction model only needs to calculate a small number of coded data classes, which can then be decoded by the decoder into spectral emissivity corresponding to the color value. In this way, the detection speed of the spectral reconstruction model can be effectively improved, and the detection accuracy and recognition precision of the spectral reconstruction model can be further improved. Furthermore, since the detection network has the function of clustering and sorting multiple color values, the color values can be divided and sorted into several ordered color clusters, and the detection network can be trained based on these ordered color clusters. Therefore, the accuracy of the spectral reconstruction model can be further improved, and the predictive ability of the spectral reconstruction model for the spectral reflectance of colors not in the training set can be optimized. Attached Figure Description
[0016] Figure 1 This is a block diagram of the spectral detection device in an embodiment of the present invention;
[0017] Figure 2 This is a schematic diagram of the detection bar used as a target image in an embodiment of the present invention;
[0018] Figure 3 This is a block diagram of the spectral reconstruction model in an embodiment of the present invention;
[0019] Figure 4 This is a flowchart of the initial detection network training process in an embodiment of the present invention;
[0020] Figure 5 This is a flowchart illustrating the operation of the spectral detection device in this embodiment of the invention;
[0021] Figure 6 This is a comparison table of experimental results in the embodiments of the present invention. Detailed Implementation
[0022] To make the technical means, creative features, objectives and effects of the present invention easy to understand, the spectral detection device of the present invention will be specifically described below in conjunction with embodiments and accompanying drawings.
[0023] <Example>
[0024] Figure 1 This is a block diagram of a spectral detection device according to an embodiment of the present invention.
[0025] like Figure 1 As shown, the spectral detection device 10 includes an image acquisition unit 1, a color value extraction unit 2, a spectral reconstruction unit 3, and an output unit 4.
[0026] The image acquisition unit 1 is used to acquire a target image of a target object, wherein the target image has at least one color block to be measured.
[0027] In this embodiment, the image acquisition unit 1 is a scanning head, the target object is a printed product, and the target image is a scanned image obtained by the scanning head scanning the printed product.
[0028] Figure 2 This is a schematic diagram of the detection bar used as a target image in an embodiment of the present invention.
[0029] like Figure 2 As shown, the printed product 11 includes a printed image area 5 and a detection strip 6. The detection strip 6 contains a color strip 7 composed of color blocks of the same color and a color strip 8 composed of multiple color blocks of different colors. Each color block to be tested is a rectangular color block. In this embodiment, the width of the rectangular color block is 0.2 mm, and the width of the detection strip 6 is 1 mm.
[0030] Color value extraction unit 2 is used to extract the color value of the color block to be tested as the color value to be tested.
[0031] In this embodiment, the color value is an RGB value. The color value extraction unit 2 first determines the pixel point in the scanned image corresponding to each color block to be tested, and then reads the RGB value of the pixel point to use as the color value to be tested for each color block to be tested.
[0032] The spectral reconstruction unit 3 is used to reconstruct the corresponding spectral reflectance based on the color value to be measured.
[0033] like Figure 3 As shown, the spectral reconstruction unit 3 includes a pre-trained spectral reconstruction model 31, which includes an autoencoder 21 and a detection network 22. Specifically:
[0034] The autoencoder 21 has an encoder for encoding spectral reflectance with multiple bands into a predetermined number of coded data and a corresponding decoder for decoding the coded data into spectral reflectance.
[0035] In this embodiment, the spectral reflectance includes spectral values corresponding to 31 bands, and the predetermined number of encoded data is preferably 3 or 8 classes, that is, the encoder can encode 31 spectral values into 3 or 8 classes of encoded data.
[0036] The detection network 22 is constructed based on the least squares support vector regression model and is obtained through pre-training.
[0037] In this embodiment, the least squares support vector regression model includes a radial basis kernel function.
[0038] Figure 4 This is a flowchart of the training process for the initial detection network in this embodiment.
[0039] like Figure 4 As shown, the training process of the initial detection network includes steps S1 to S6, as detailed below:
[0040] Step S1: Obtain a standard image obtained by scanning a standard printing template by the image acquisition unit 1. The standard image contains multiple standard color blocks.
[0041] In this embodiment, the standard printing template can be a Munsell color chart, an NCS (Natural Colour System) color chart, etc.
[0042] Step S2: Obtain the color value of the standard color block as the standard color value and the corresponding spectral reflectance as the standard spectral reflectance.
[0043] In this embodiment, the standard color value is obtained by the color value extraction unit 2. The specific extraction method is the same as the method used by the color value extraction unit 2 to extract the color value to be measured, and will not be described again here. The standard spectral reflectance is the spectral reflectance obtained by detecting each color patch with a spectrometer.
[0044] Step S3: Cluster the RGB values (i.e., standard color values) in the RGB color space so that multiple standard color values are divided into several color clusters.
[0045] Step S4: Sort the standard color values in each color cluster according to their size to form several corresponding ordered color clusters.
[0046] In this embodiment, step S4 includes the following sub-steps:
[0047] Step S4-1: Sort each color cluster sequentially according to its RGB value to form an RGB value sequence;
[0048] Step S4-2: Extract a predetermined number of RGB values from the RGB value sequence at predetermined intervals, and form an ordered color cluster from the extracted RGB values.
[0049] Step S5: Input the standard spectral reflectance corresponding to each standard color value into the autoencoder 21 in sequence to obtain the corresponding encoded data as standard encoded data.
[0050] Step S6: According to the order of the standard color values in each ordered color cluster, the initial detection network is trained sequentially based on the standard color value and the corresponding standard encoding data until the predetermined conditions are met and the trained detection network 22 is obtained.
[0051] In this embodiment, the ordered color clusters and standard encoded data are divided into two datasets of equal size. One dataset is used for training the detection network 22, i.e., the training dataset; the other dataset is used for testing the trained detection network 22, i.e., the test dataset.
[0052] In this embodiment, the predetermined condition is to determine whether the detection effect of the detection network meets a predetermined threshold, specifically:
[0053] First, the trained detection network is tested using a test dataset.
[0054] Next, determine whether the test result meets the predetermined threshold. If the result is yes, the detection network is successfully trained. If the result is no, repeat step S6, reselect the training dataset, and retrain the detection network.
[0055] In this embodiment, the test result is the value of the average color difference deltaE00, and the predetermined threshold is 0.10.
[0056] To better illustrate the effect of the spectral reconstruction model in this embodiment, the color values to be tested in the training dataset and the test dataset are input into the successfully trained detection network, and the predicted encoded data corresponding to each color value to be tested is output. The autoencoder 21 then decodes the data to obtain the predicted value of the spectral reflectance corresponding to the color value to be tested. The evaluation parameters are calculated using the predicted value and the actual value of the spectral reflectance corresponding to each color value to be tested.
[0057] The evaluation parameters include root mean square error, average color difference, maximum color difference, and one or more fitting evaluation factors.
[0058] Figure 6 This is a comparison table of experimental results in the embodiments of the present invention.
[0059] The experimental data in this embodiment are as follows: Figure 6As shown, for the training set data, the calculated root mean square error is 0.0065, the average color difference deltaE00 is 0.0799, the maximum color difference deltaE00 is 0.510, and the evaluation factor is 0.9995; for the test set data, the calculated root mean square error is 0.0137, the average color difference deltaE00 is 0.10, the maximum color difference is 1.1, and the fitting evaluation factor is 0.9980. In this application scenario, the maximum allowable color difference deltaE00 is generally 2.5, therefore, the trained detection network is deemed qualified.
[0060] Furthermore, according to the experimental results comparison table, it can be clearly seen that, compared with the experimental data output by the prior art, the spectral reconstruction model 31 provided in this embodiment has lower root mean square error, average color difference, and maximum color difference, whether it is the training dataset or the test dataset. Therefore, the recognition accuracy of the spectral reconstruction model 31 is higher.
[0061] It should be noted that the calculation formulas for root mean square error, average color difference, maximum color difference, and fitting evaluation factor can all adopt commonly used calculation formulas in the prior art, and will not be elaborated on in this embodiment.
[0062] Through the above process, a spectral reconstruction model 31 can be trained, which can accurately reconstruct the spectral reflectance corresponding to the color value of the color patch to be tested.
[0063] Output unit 4 is used to output the color value and corresponding spectral reflectance of the color block to be tested as the detection result. In this embodiment, output unit 4 can output the detection result to a system used to detect whether there are problems with printed products based on spectral data.
[0064] Figure 5 This is a flowchart of the operation of the spectral detection device 10 according to an embodiment of the present invention.
[0065] like Figure 5 As shown, the working process of the spectral detection device 10 includes the following steps:
[0066] Step A1: The image acquisition unit 1 scans the target object with the scanning head to obtain the corresponding target image, and then proceeds to step A2;
[0067] Step A2: The color value extraction unit 2 extracts the color value of the color block to be tested as the color value to be tested, and then proceeds to step A3;
[0068] Step A3: Input the color value to be measured into the spectral reconstruction unit 3, and then proceed to step A4;
[0069] Step A4: The spectral reconstruction model 31 clusters the color values to be measured through the detection network 22 and sorts them according to the standard color value size to obtain several ordered color clusters, and then proceeds to step A5;
[0070] Step A5: The autoencoder 21 encodes the spectral reflectance into a predetermined number of encoded data and decodes the encoded data to obtain the spectral reflectance corresponding to each ordered color cluster, and then proceeds to step A6.
[0071] In step A6, the output unit 4 outputs the reconstructed spectral reflectance corresponding to the color value to be measured, and then enters the end state.
[0072] In this embodiment, the above process can also be simplified to an image acquisition step (i.e., step A1), a color value extraction step (i.e., step A2), and a spectral reconstruction step (i.e., steps A3 to A5), thereby forming a method capable of spectral detection of the color of a target object.
[0073] Functions and effects of the embodiments
[0074] According to the spectral detection device provided in this embodiment, the color value of the color patch to be tested is extracted from the target image scanned by the image acquisition unit by the color value extraction unit, and then spectral reconstruction is performed by the spectral reconstruction unit. This spectral reconstruction unit has a pre-trained spectral reconstruction model, which includes a detection network and an autoencoder. The autoencoder has an encoder for encoding the spectral reflectance of multiple bands into a predetermined number of coded data classes and a corresponding decoder for decoding the coded data into spectral reflectance. Therefore, the spectral reconstruction model only needs to calculate a small number of coded data classes, which can then be decoded by the decoder into spectral emissivity corresponding to the color value. In this way, the detection speed of the spectral reconstruction model can be effectively improved, and the detection accuracy and recognition precision of the spectral reconstruction model can be further enhanced. Furthermore, since the detection network has the function of clustering and sorting multiple color values, the color values can be divided and sorted into several ordered color clusters, and the detection network can be trained based on these ordered color clusters. Therefore, the accuracy of the spectral reconstruction model can be further improved, and the predictive ability of the spectral reconstruction model for the spectral reflectance of colors not in the training set can be optimized.
[0075] In this embodiment, since the autoencoder 21 can divide the 31 types of spectral data into 3 or 8 categories, it greatly reduces the number of data that the spectral reconstruction model needs to process during classification, making the final structure of the spectral reconstruction model simpler and improving the classification speed. Furthermore, when 8 categories are preferred, the classification effect can be made as good as possible while ensuring speed; when classifying into 3 categories, the speed is the fastest, but the performance requirements of the decoder are higher, which can easily affect the accuracy.
[0076] In this embodiment, since the detection strip 6 only needs to be 1mm (the width of each color patch to be tested is 0.2mm), after being scanned by the spectral detection device 10 of this embodiment, the corresponding color value to be tested can be extracted, and the corresponding spectral reflectance can be accurately reconstructed by the spectral detection device 10. Therefore, for printed products, only a very small detection strip is needed to complete the detection. In contrast, this avoids the need for a detection strip of at least 3cm required in traditional detection, which needs to be cut and discarded after the detection is completed, which is very wasteful. The spectral detection device 10 of this embodiment can effectively solve this problem, achieving both energy saving and environmental protection, as well as high efficiency and accuracy.
[0077] The above embodiments are only used to illustrate specific implementations of the present invention, and the present invention is not limited to the scope of the description of the above embodiments.
[0078] In the above embodiments, the product being tested is a paper print. In other embodiments of the present invention, it can also be used for any other form or material, such as liquids or packaging, that requires color detection.
[0079] In the above embodiments, the predetermined number of data types is 3 or 8. In other embodiments of the present invention, the predetermined number of data types can also be set to any number of types between 3 and 12.
[0080] In the above embodiments, the width of each color block to be tested is 0.2 mm, and the width of the detection strip is 1 mm. In other embodiments of the present invention, the width of each color block to be tested is between 0.2 mm and 2 mm, and the width of the detection strip varies with the width of the color block to be tested.
[0081] In the above embodiment, the predetermined condition is to determine whether the test result is less than a predetermined threshold. As an alternative, the predetermined condition can also be to determine whether the number of training rounds of the detection network is more than 1000 rounds.
Claims
1. A spectral detection device for performing spectral detection on the color of a target object, characterized in that, include: An image acquisition unit is used to acquire a target image obtained by image acquisition of the target object, wherein the target image has at least one color block to be tested; The color value extraction unit extracts the color value of the color block to be tested as the color value to be tested. The spectral reconstruction unit has a pre-trained spectral reconstruction model, which is used to reconstruct the corresponding spectral reflectance based on the color value to be measured. The output section is used to output the spectral reflectance; The spectral reflectance includes multiple spectral bands and spectral values for each spectral band. The spectral reconstruction model has an autoencoder and a detection network. The autoencoder has an encoder for encoding the spectral reflectance having multiple bands into a predetermined number of coded data types, and a corresponding decoder for decoding the coded data back into the spectral reflectance. The detection network was obtained in advance through the following training steps: Step S1: Obtain a standard image obtained by the image acquisition unit scanning the standard printing template. The standard image contains multiple standard color blocks. Step S2: Obtain the color value of the standard color block as the standard color value and the corresponding spectral reflectance as the standard spectral reflectance; Step S3: Cluster the standard color values and divide the multiple standard color values into several color clusters; Step S4: Sort the standard color values in each color cluster according to their size to form several ordered color clusters. Step S5: Input the standard spectral reflectance corresponding to each standard color value into the autoencoder in sequence to obtain the corresponding encoded data as standard encoded data; Step S6: Following the order of the standard color values in each ordered color cluster, the detection network is trained sequentially based on the standard color value and the corresponding standard encoding data until a predetermined condition is met and a trained detection network is obtained. The trained detection network can obtain corresponding predictive encoded data based on the color value to be tested, and the predictive encoded data can be decoded by the decoder into the spectral reflectance corresponding to the color value to be tested.
2. The spectral detection device according to claim 1, characterized in that: in, The spectral reflectance has 31 bands, and the predetermined number of classes is 3 to 12.
3. The spectral detection device according to claim 2, characterized in that: in, The predetermined quantity can be 3 or 8 categories.
4. The spectral detection device according to claim 1, Its features are: The predetermined condition is to determine whether the detection effect of the detection network meets a predetermined threshold. The trained detection network is tested using a predetermined test dataset; Determine whether the test result meets a predetermined threshold. If the result is yes, the detection network is successfully trained. If the result is no, repeat step S6 to train the detection network again.
5. The spectral detection device according to claim 1, characterized in that: in, The color patch to be tested is a rectangular color patch with a width between 0.2 mm and 2 mm.
6. The spectral detection device according to claim 1, characterized in that: in, The color value is an RGB value. Step S4 includes the following sub-steps: Step S4-1: Sort each color cluster sequentially according to the size of its RGB values to form an RGB value sequence; Step S4-2: Extract a predetermined number of RGB values from the RGB value sequence at predetermined intervals, and combine the extracted RGB values into the ordered color cluster.
7. The spectral detection device according to claim 1, characterized in that: in, The image acquisition unit is a scanning head used to scan the target object to obtain the target image.
8. A spectral detection method, implemented using the spectral detection device as described in any one of claims 1-7, for performing spectral detection on the color of a target object, characterized in that, include: The image acquisition step involves acquiring a target image obtained by image acquisition of the target object, wherein the target image contains at least one color patch to be tested. The color value extraction step involves extracting the color value of the color block to be tested as the color value to be tested. The spectral reconstruction step involves inputting the color value to be measured into a pre-trained spectral reconstruction model, which then reconstructs the spectral reflectance corresponding to the color value to be measured. The output section is used to output the spectral reflectance; The spectral reconstruction model includes an autoencoder and a detection network. The spectral reflectance includes multiple bands and spectral values for each band. The autoencoder has encoded data for encoding the spectral reflectance having multiple bands into a predetermined number of classes, and a corresponding decoder for decoding the encoded data into the spectral reflectance. The detection network was obtained in advance through the following training steps: Step S1: Obtain a standard image obtained by the image acquisition unit scanning the standard printing template. The standard image contains multiple standard color blocks. Step S2: Obtain the color value of the standard color block as the standard color value and the corresponding spectral reflectance as the standard spectral reflectance; Step S3: Cluster the standard color values and divide the multiple standard color values into several color clusters; Step S4: Sort the standard color values in each color cluster according to their size to form several ordered color clusters. Step S5: Input the standard spectral reflectance corresponding to each standard color value into the autoencoder in sequence to obtain the corresponding encoded data as standard encoded data; Step S6: Following the order of the standard color values in each ordered color cluster, the initial detection network is trained sequentially based on the standard color value and the corresponding standard encoding data until a predetermined condition is met and the trained detection network is obtained. The trained detection network can obtain corresponding predictive encoded data based on the color value to be tested, and the predictive encoded data can be decoded by the decoder into the spectral reflectance corresponding to the color value to be tested.