Color difference correction and chromaticity detection method and system based on DDGS

Abstract

The present invention discloses a method and system based on DDGS color difference correction and color detection, the method comprising: photographing a DDGS feed to be detected to obtain a DDGS image; performing image preprocessing on the DDGS image to obtain a DDGS image to be detected; constructing and training a DDGS color difference correction neural network model; inputting the DDGS image to be detected into the DDGS color difference correction neural network model to obtain three-channel RGB values ​​in the DDGS image to be detected, and generating a DDGS image after color difference correction; converting the three-channel RGB values ​​in the DDGS image after color difference correction to the LAB color domain for color grading, and outputting the grading of the DDGS feed to be detected. The present invention determines the basic chromaticity level and the biased chromaticity level by calculating the color difference between the DDGS image after color difference correction and samples of different levels of the standard color card, and finally combines the color difference of the two levels to give a more accurate color grading.

Description

Technical Field

[0001] The invention relates to the technical field of DDGS detection, and in particular to a method and system for color difference correction and chromaticity detection based on DDGS. Background Art

[0002] DDGS feed is a byproduct of ethanol fermentation using corn as raw material. Its surface color is closely related to its nutritional content and is one of the important factors affecting its market value. At present, laboratories mainly use cutting-edge instruments such as spectrometers and colorimeters to measure the spectral information and colorimetric information of DDGS to accurately grade DDGS. Although this method has the highest accuracy, its high cost and complex operating procedures make it difficult to be widely used in large-scale production of enterprises. Therefore, most companies still use the visual method to grade DDGS feed by comparing color cards. This method has several significant problems: (1) The core of the visual method relies on the operator's color sensitivity and experience. The color sensitivity of people varies greatly, so there is a large degree of subjectivity and error. Even experienced operators may make misjudgments due to factors such as light, environment, and fatigue. (2) The visual method cannot accurately grade DDGS. It can only correspond to the level with the closest color, and cannot achieve accurate quantitative evaluation. This means that products with large color differences may be classified into the same level. This not only leads to inaccurate grading results, but also makes it difficult for different companies to form a unified standard for DDGS feed quality, which brings great inconvenience to market transactions. (3) The visual method is a manual operation technology that cannot meet the needs of real-time product monitoring. If the production environment changes and causes the DDGS color to decrease, it is impossible to find and solve the problem in time, which may lead to fluctuations and losses in product quality. Therefore, the development of a scientific, quantifiable, and automated color evaluation system is urgently needed in the current DDGS feed market. Summary of the invention

[0003] Technical purpose: In view of the defects in the prior art, the present invention discloses a method and system for color difference correction and chromaticity detection based on DDGS. By calculating the color difference between the DDGS image after color difference correction and samples of different levels of the standard color card, the basic chromaticity level and the biased chromaticity level are determined, and finally a more accurate color grading is given by combining the color difference of the two levels.

[0004] Technical solution: In order to achieve the above technical objectives, the present invention adopts the following technical solution.

[0005] A method for color difference correction and colorimetry detection based on DDGS, the method comprising:

[0006] For the DDGS feed to be tested, the DDGS image is obtained by photographing it through a semi-enclosed box structure;

[0007] Perform image preprocessing on the DDGS image to obtain the DDGS image to be detected;

[0008] Constructing and training a DDGS color difference correction neural network model; inputting the DDGS image to be detected into the DDGS color difference correction neural network model, obtaining the three-channel RGB values ​​in the DDGS image to be detected, and generating a DDGS image after color difference correction;

[0009] The three-channel RGB values ​​in the DDGS image after color difference correction are converted to the LAB color gamut for color grading, and the grade of the DDGS feed to be tested is output.

[0010] Preferably, the DDGS image is preprocessed to obtain the DDGS image to be detected, including: performing reflectance gain processing on the DDGS image, and then performing cropping and blurring processing to remove texture features.

[0011] Preferably, the reflectance gain processing includes: converting the photographed DDGS image to the HLS color gamut, multiplying the brightness value L by a gain coefficient ρ; the gain coefficient is obtained by photographing a DDGS feed sample and a standard white card, blurring them respectively, converting the blurred standard white card and the photographed DDGS feed sample to the HLS color gamut, and calculating the ratio between the maximum values ​​of L as the gain coefficient.

[0012] Preferably, constructing and training a DDGS color difference correction neural network model includes:

[0013] Construct a sample set;

[0014] A DDGS color difference correction neural network model is constructed, including an input layer, two residual structures and an output layer connected in sequence; the residual structure includes: a fully connected layer 1, a BN layer 1, a PReLU activation function 1, a fully connected layer 2, a BN layer 2, a PReLU activation function 2, and a fully connected layer 3 connected in sequence; the fully connected layer 1 and the fully connected layer 3 are connected to form a residual structure through a jump connection; the input layer inputs the three-channel values ​​of R, G, and B; the output layer outputs the three-channel values ​​of R, G, and B through a Sigmoid activation function; the value range is 0 to 1;

[0015] The DDGS color difference correction neural network model was obtained by training: the constructed DDGS color difference correction neural network model was trained by using 5-fold cross validation in combination with the sample set. The input was the RGB value of the area near the center point of each color block in the standard color card image, the label was the RGB value of the corresponding color block of the standard color card, and the output was the predicted three-channel RGB value; the trained DDGS color difference correction neural network model was obtained.

[0016] Preferably, constructing the sample set includes the following steps:

[0017] Step S1, photographing a standard color card through a semi-enclosed box structure to obtain a standard color card image; wherein a stable, uniform, high color rendering index light source is provided in the semi-enclosed box structure to facilitate photographing;

[0018] Step S2, extracting the coordinates of the center point of the color block from the standard color card image;

[0019] Step S3: Establish a mapping relationship between the standard color card image and the standard color card. For the RGB values ​​of points in the area near the center point of each color block in the standard color card image, the corresponding label is the RGB value of the corresponding color block of the standard color card.

[0020] Preferably, the process of color grading includes:

[0021] The three-channel RGB values ​​in the color-aberration-corrected DDGS image are converted to the LAB color gamut to calculate the color difference and obtain the basic classification; the color-aberration-corrected DDGS image is converted from the RGB color gamut to the LAB color gamut, and then the average of each channel L, A, and B is taken, and the color difference CIEDE2000 is calculated with the LAB value of each level of samples on the standard DDGS colorimetric card, and the classification with the smallest color difference is selected as the basic classification;

[0022] The three-channel RGB values ​​in the DDGS image after color difference correction are converted to the LAB color gamut, and the brightness value L is compared with the basic grading to obtain the biased grading, where the brightness value L is the average value of the L value after conversion to the LAB color gamut;

[0023] The final classification of the DDGS feed to be tested is calculated based on the basic classification and the biased classification.

[0024] Preferably, the calculation formula for the final classification includes:

[0025]

[0026] Among them level 1 For basic classification, level 2 For bias classification, DE 1 is the color difference between the color-corrected image and the basic grading, DE 2 It is the color difference between the image after color correction and the bias grading.

[0027] A DDGS-based color difference correction and colorimetry detection system, used to implement any of the above-mentioned DDGS-based color difference correction and colorimetry detection methods, comprising:

[0028] The detection and shooting module is used to obtain the DDGS image of the DDGS feed to be tested through a semi-enclosed box structure. The detection and shooting module is used to maintain the consistency of the testing environment. It adopts a semi-enclosed box structure, and uses a stable, uniform, and high color rendering index light source inside the box.

[0029] An image preprocessing module is used to perform image preprocessing on the DDGS image to obtain a DDGS image to be detected;

[0030] The color difference correction module is used to construct and train a DDGS color difference correction neural network model; input the DDGS image to be detected into the DDGS color difference correction neural network model, obtain the three-channel RGB values ​​in the DDGS image to be detected, and generate a DDGS image after color difference correction;

[0031] The DDGS color detection module is used to convert the three-channel RGB values ​​in the DDGS image after color difference correction into the LAB color domain for color grading, and output the grade of the DDGS feed to be detected.

[0032] Beneficial effects:

[0033] 1. The present invention is based on a neural network model, and a mapping relationship is established between the photographed color card and the standard color card, and a neural network model for color correction is trained to achieve high-quality chromaticity correction;

[0034] 2. The present invention determines the basic chromaticity level and the biased chromaticity level by calculating the color difference between the DDGS image after color difference correction and samples of different levels of the standard colorimetric card, and finally combines the color difference of the two levels to give a more accurate color grading, thereby realizing a more scientific and accurate grading scheme. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 is a flow chart of the method of the present invention;

[0036] Figure 2 is a system block diagram of the present invention;

[0037] Figure 3 It is a schematic diagram of the box structure of the present invention;

[0038] Figure 4 This is a schematic diagram of a standard color card;

[0039] Figure 5 Schematic diagram of the rectangular frame output after edge detection;

[0040] Figure 6 Schematic diagram of the neural network model structure for DDGS color difference correction. DETAILED DESCRIPTION

[0041] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0042] Example:

[0043] As attached Figure 1 As shown, a method for color difference correction and chromaticity detection based on DDGS of the present invention comprises the following steps:

[0044] 1. For the DDGS feed to be tested, the DDGS image is obtained by shooting through a semi-enclosed box structure; as shown in the attached Figure 3 As shown, the DDGS feed to be tested is continuously conveyed through the conveying port by a conveyor belt, and a semi-enclosed box structure is set on the conveyor belt, as shown in the attached Figure 3 As shown, the semi-enclosed box structure has a built-in light source and a camera, which are used to photograph the DDGS feed to be tested to obtain a DDGS image. A constant light source will cause heat and brightness attenuation. The present invention adopts a trigger design, and a shooting interval is preset, such as shooting once per minute. When the shooting is triggered, the light source is turned on, and after the light source is stable, a photo is taken, and then the light source is turned off. The photo is transmitted to the server for further processing and a corresponding classification is given. In other words, the subsequent steps of this method can be implemented on the server.

[0045] 2. Preprocessing the DDGS image to obtain the DDGS image to be detected; the image preprocessing includes: performing reflectance gain processing on the DDGS image, and then performing cropping and blurring processing to remove texture features;

[0046] In order to solve the problem that the color of the captured picture is dark due to the low reflectivity of the object, the present invention performs reflectance gain processing on the captured DDGS image. Specifically, the captured DDGS image is converted from the RGB color gamut to the HLS color gamut, and the brightness value L is multiplied by a gain coefficient ρ; after the adjustment is completed, it is converted back to the RGB color gamut to achieve color gamut conversion, and the DDGS image after reflectance gain adjustment is obtained; this process can be implemented by a program; the DDGS image after reflectance gain adjustment is cropped and blurred to remove texture features and retain color features to obtain a DDGS image to be detected, and improve the classification accuracy, specifically by cropping the central area of ​​the captured image, and performing three consecutive Gaussian blurs and average pooling to obtain a blurred image;

[0047] The gain coefficient is obtained by photographing a DDGS feed sample and a standard white card, blurring them respectively, converting the blurred standard white card and the photographed DDGS feed sample to the HLS color gamut, and calculating the ratio between the maximum values ​​of L as the gain coefficient;

[0048] 3. Construct and train a neural network model for DDGS color difference correction; the process includes:

[0049] Construct a sample set; including the following steps:

[0050] Step S1, photographing a standard color card through a semi-enclosed box structure to obtain a standard color card image; wherein a stable, uniform, high color rendering index light source is provided in the semi-enclosed box structure to facilitate photographing;

[0051] Step S2, extracting the coordinates of the center point of the color block from the standard color card image; comprising:

[0052] Read the grayscale image of the standard color card image, perform edge detection through the Candy operator, output the edges of several color blocks on the standard color card, and then use rectangle detection to frame the color block with the smallest rectangular frame, output the coordinates of the four vertices of the rectangular frame, and obtain the coordinates of the center point through the four vertex coordinates.

[0053] As attached Figure 4 As shown, attached Figure 4 The standard color card is a schematic diagram, which is 8 rows and 8 columns of different color blocks; read the grayscale image of the standard color card image, perform edge detection through the Candy operator, and output the edges of several color blocks on the standard color card. Figure 5 The following is a schematic diagram of the rectangular box output after edge detection. At this time, some color blocks are not detected, so the center point cannot be output. We cluster the center of each row and the center of each column through other detected center points. The output result is the center line of each row of 8 rows of color blocks and the center line of each column of 8 columns of color blocks. The midpoint of each color block is the arrangement and combination of each center line.

[0054] Step S3, establishing a mapping relationship between the standard color card image and the standard color card, and for the RGB values ​​of the points in the area near the center point of each color block in the standard color card image, the corresponding label is the RGB value of the corresponding color block of the standard color card;

[0055] As attached Figure 6 As shown, a DDGS color difference correction neural network model is constructed, including an input layer, two residual structures and an output layer connected in sequence; Figure 6The first column represents the input layer, and the input is the three-channel values ​​of R, G, and B. The three-channel values ​​of R, G, and B are normalized to a range of 0 to 1 by dividing them by 255; the last column is the output layer, and the three-channel values ​​of R, G, and B are output through the Sigmoid activation function; the value range is 0 to 1; there are two residual structures between the input layer and the output layer, that is, the neural networks of the two adjacent yellow areas form a residual structure through skip-connection; the output of the first residual structure is the input of the second residual structure;

[0056] The residual structure includes: a fully connected layer 1, a BN layer 1 (Batch Normalization layer), a PReLU activation function 1, a fully connected layer 2, a BN layer 2, a PReLU activation function 2, and a fully connected layer 3 are sequentially connected; the fully connected layer 1 and the fully connected layer 3 are connected through a jump connection to form a residual structure; wherein the fully connected layer 1 is Figure 6 The neurons in the second column, BN layer 1, PReLU activation function 1, and fully connected layer 2 are collectively displayed as the third column of neurons, and BN layer 2, PReLU activation function 2, and fully connected layer 3 are the fourth column of neurons; the network structures of the two residual structures are the same and will not be repeated here.

[0057] The DDGS color difference correction neural network model is obtained by training: the constructed DDGS color difference correction neural network model is trained by using a 5-fold cross validation in combination with the sample set, the input is the RGB value of the area near the center point of each color block in the standard color card image, the label is the RGB value of the corresponding color block of the standard color card, and the output is the predicted three-channel RGB value; the trained DDGS color difference correction neural network model is obtained;

[0058] 4. Input the DDGS image to be tested into the DDGS color difference correction neural network model, obtain the three-channel RGB values ​​in the DDGS image to be tested, and generate a DDGS image after color difference correction. The DDGS image after color difference correction is an RGB image; convert the three-channel RGB values ​​in the DDGS image after color difference correction to the LAB color gamut for color grading, and output the grade of the DDGS feed to be tested.

[0059] The process of color grading includes:

[0060] The three-channel RGB values ​​of the DDGS image after color difference correction were converted to the LAB color gamut to calculate the color difference and obtain the basic classification. The DDGS image after color difference correction was converted from the RGB color gamut to the LAB color gamut, and then the average of the L, A, and B channels was taken, and the color difference CIEDE2000 was calculated with the LAB values ​​of each level of samples on the standard DDGS colorimetric card, and the classification with the smallest color difference was selected as the basic classification.

[0061] The three-channel RGB values ​​in the DDGS image after color difference correction are converted to the LAB color gamut, and the brightness value L is compared with the basic grading to obtain the bias grading, wherein the brightness value L is the average value of the L value after conversion to the LAB color gamut; the calculation process of the bias grading is as follows: if the three-channel RGB values ​​in the DDGS image after color difference correction are converted to the LAB color gamut, the average value of the L channel is taken, recorded as the first L value, and the L value of the basic grading is obtained as the second L value. If the first L value is greater than the second L value, the bias grading is the basic grading -1; if the first L value is less than the second L value, the bias grading is the basic grading +1; if the first L value is equal to the second L value, the bias grading is the basic grading;

[0062] That is to say, in the present invention, the basic classification is obtained by comparing the average values ​​of L, A, and B with the color card, and the biased classification is obtained by comparing the average value of L with the basic classification; the final classification of the DDGS feed to be tested is calculated according to the basic classification and the biased classification; the relative position is calculated by the color difference between the DDGS after color calibration and the basic classification and the biased classification to give a more accurate chromaticity classification of DDGS; the calculation formula for the final classification is as follows:

[0063]

[0064] Among them level 1 For basic classification, level 2 For bias classification, DE 1 is the color difference between the color-corrected image and the basic grading, DE 2 It is the color difference between the image after color correction and the bias classification. The color difference is calculated by CIEDE2000 formula, and the calculation formula is as follows:

[0065]

[0066] K L , K C , K H are the gain parameters of brightness, chroma, and hue, respectively, which are set to 1 in this paper;

[0067] Assume that the two colors to be calculated for color difference are (L 1 , a 1 , b 1 ) and (L 2 , a 2 , b 2 ), first calculate:

[0068]

[0069] in,

[0070]

[0071] in, C′ ab，1 and C′ ab，2 Mean;

[0072] From this we can calculate:

[0073]

[0074] Where Δh ab ′=h′ ab，1 -h′ ab，2 ;

[0075] Final calculation:

[0076]

[0077] in, For L 1 ′ and L 2 The mean of ′

[0078]

[0079] in, h′ ab，1 and h′ ab，2 The mean of .

[0080] In some embodiments of the present invention, the relative position is calculated by the color difference between the calibrated DDGS and the basic classification and the biased classification to give a more accurate DDGS classification. For example, if the basic classification is 3 and the biased classification is 4, the color difference with the basic classification is 10 and the color difference with the biased classification is 15, then the final classification is 3+10 / (10+15)=3.4. If the biased classification is 2, then the final classification is 3-10 / (10+15)=2.6.

[0081] As attached Figure 2 As shown, the present invention also discloses a DDGS-based color difference correction and chromaticity detection system, which is used to implement the above-mentioned DDGS-based color difference correction and chromaticity detection method, including:

[0082] The detection and shooting module is used to obtain the DDGS image of the DDGS feed to be tested through a semi-enclosed box structure. The detection and shooting module is used to maintain the consistency of the testing environment. It adopts a semi-enclosed box structure, and uses a stable, uniform, and high color rendering index light source inside the box.

[0083] An image preprocessing module is used to perform image preprocessing on the DDGS image to obtain a DDGS image to be detected;

[0084] The color difference correction module is used to construct and train a DDGS color difference correction neural network model; input the DDGS image to be detected into the DDGS color difference correction neural network model, obtain the three-channel RGB values ​​in the DDGS image to be detected, and generate a DDGS image after color difference correction;

[0085] The DDGS color detection module is used to convert the three-channel RGB values ​​in the DDGS image after color difference correction into the LAB color domain for color grading, and output the grade of the DDGS feed to be detected.

[0086] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for color difference correction and colorimetry detection based on DDGS, characterized in that: The method comprises: For the DDGS feed to be tested, the DDGS image is obtained by photographing it through a semi-enclosed box structure; Perform image preprocessing on the DDGS image to obtain the DDGS image to be detected; Constructing and training a DDGS color difference correction neural network model; inputting the DDGS image to be detected into the DDGS color difference correction neural network model, obtaining the three-channel RGB values ​​in the DDGS image to be detected, and generating a DDGS image after color difference correction; The three-channel RGB values ​​in the DDGS image after color difference correction are converted to the LAB color space for color grading, and the grade of the DDGS feed to be tested is output; wherein the color grading process includes: The three-channel RGB values ​​in the color-aberration-corrected DDGS image are converted to the LAB color gamut to calculate the color difference and obtain the basic classification; the color-aberration-corrected DDGS image is converted from the RGB color gamut to the LAB color gamut, and then the average of each channel L, A, and B is taken, and the color difference CIEDE2000 is calculated with the LAB value of each level of samples on the standard DDGS colorimetric card, and the classification with the smallest color difference is selected as the basic classification; The three-channel RGB values ​​in the DDGS image after color difference correction are converted to the LAB color gamut, and the brightness value L is compared with the basic grading to obtain the biased grading, where the brightness value L is the average value of the L value after conversion to the LAB color gamut; The final classification of the DDGS feed to be tested is calculated based on the basic classification and the biased classification.

2. A method for color difference correction and chromaticity detection based on DDGS according to claim 1, characterized in that: The DDGS image is preprocessed to obtain the DDGS image to be detected, including: performing reflectance gain processing on the DDGS image, and then performing cropping and blurring processing to remove texture features.

3. A method for color difference correction and chromaticity detection based on DDGS according to claim 2, characterized in that: The reflectance gain processing includes: converting the photographed DDGS image to the HLS color gamut, multiplying the brightness value L by a gain coefficient ρ; the gain coefficient is obtained by photographing a DDGS feed sample and a standard white card, blurring them respectively, converting the blurred standard white card and the photographed DDGS feed sample to the HLS color gamut, and calculating the ratio between the maximum values ​​of L as the gain coefficient.

4. A method for color difference correction and chromaticity detection based on DDGS according to claim 1, characterized in that: The DDGS color difference correction neural network model was constructed and trained, including: Construct a sample set; A DDGS color difference correction neural network model is constructed, including an input layer, two residual structures and an output layer connected in sequence; the residual structure includes: a fully connected layer 1, a BN layer 1, a PReLU activation function 1, a fully connected layer 2, a BN layer 2, a PReLU activation function 2, and a fully connected layer 3 connected in sequence; the fully connected layer 1 and the fully connected layer 3 are connected to form a residual structure through a jump connection; the input layer inputs the three-channel values ​​of R, G, and B; the output layer outputs the three-channel values ​​of R, G, and B through a Sigmoid activation function; the value range is 0 to 1; The DDGS color difference correction neural network model was obtained by training: the constructed DDGS color difference correction neural network model was trained by using 5-fold cross validation in combination with the sample set. The input was the RGB value of the area near the center point of each color block in the standard color card image, the label was the RGB value of the corresponding color block of the standard color card, and the output was the predicted three-channel RGB value; the trained DDGS color difference correction neural network model was obtained.

5. A method for color difference correction and chromaticity detection based on DDGS according to claim 4, characterized in that: Building a sample set includes the following steps: Step S1, photographing a standard color card through a semi-enclosed box structure to obtain a standard color card image; wherein a stable, uniform, high color rendering index light source is provided in the semi-enclosed box structure to facilitate photographing; Step S2, extracting the coordinates of the center point of the color block from the standard color card image; Step S3: Establish a mapping relationship between the standard color card image and the standard color card. For the RGB values ​​of points in the area near the center point of each color block in the standard color card image, the corresponding label is the RGB value of the corresponding color block of the standard color card.

6. A method for color difference correction and chromaticity detection based on DDGS according to claim 1, characterized in that: The final classification is calculated using the following formula: Among them, level1 is the basic classification, level2 is the biased classification, DE1 is the color difference between the image after color correction and the basic classification, and DE2 is the color difference between the image after color correction and the biased classification.

7. A DDGS-based color difference correction and colorimetric detection system, used to implement a DDGS-based color difference correction and colorimetric detection method as described in any one of claims 1 to 6, characterized in that: include: The detection and shooting module is used to obtain the DDGS image of the DDGS feed to be tested through a semi-enclosed box structure. The detection and shooting module is used to maintain the consistency of the testing environment. It adopts a semi-enclosed box structure, and uses a stable, uniform, and high color rendering index light source inside the box. An image preprocessing module is used to perform image preprocessing on the DDGS image to obtain a DDGS image to be detected; The color difference correction module is used to construct and train a DDGS color difference correction neural network model; input the DDGS image to be detected into the DDGS color difference correction neural network model, obtain the three-channel RGB values ​​in the DDGS image to be detected, and generate a DDGS image after color difference correction; The DDGS color detection module is used to convert the three-channel RGB values ​​in the DDGS image after color difference correction into the LAB color domain for color grading, and output the grade of the DDGS feed to be detected.

Images