A donkey-hide gelatin quality detection method and system based on image recognition

Abstract

The present invention relates to the field of image recognition technology, and specifically to a donkey-hide gelatin quality detection method and system based on image recognition. First, a time-series image data set of different stages in the donkey-hide gelatin production process is obtained through a multispectral camera, including boiling, filtering, concentration, and cooling and forming. Secondly, a donkey-hide gelatin image abnormal area recognition model is constructed to process the obtained time-series image data and identify the abnormal areas in the images of each stage. Then, based on the abnormal areas detected in the images of each stage, a donkey-hide gelatin image abnormal area detection model is constructed to compare and analyze the changes in the abnormal areas of the images, obtain the impurities abnormality in the donkey-hide gelatin images, and thus obtain the second area abnormal change coefficient. At the same time, through a multi-parameter fusion technology, the abnormal change coefficients of different stages are integrated to obtain the donkey-hide gelatin abnormality coefficient. Finally, according to the donkey-hide gelatin abnormality coefficient, the donkey-hide gelatin is classified into first-class, second-class, and third-class, providing a quality assessment mechanism for the quality of donkey-hide gelatin.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and particularly to a donkey-hide gelatin quality detection method and system based on image recognition. Background Art

[0002] Donkey-hide gelatin is a traditional Chinese medicinal material, which is widely used in the fields of health care and medicine due to its nourishing and medical effects. With the increasing market demand, the production scale of donkey-hide gelatin is expanding day by day, and the requirements for the detection of its product quality are also increasing. The quality of donkey-hide gelatin is mainly evaluated through characteristics such as appearance, color, gloss, transparency, bubbles, and impurities. The existing donkey-hide gelatin quality detection methods mainly rely on sampling judgment of finished donkey-hide gelatin and edge detection algorithms to detect leakage strips on the surface of donkey-hide gelatin. The detection process depends on identifying the surface characteristics of finished donkey-hide gelatin. This method not only cannot regulate the production process, but also has low efficiency, low detection accuracy, and is easily affected by the subjective judgment of the detection personnel, resulting in poor stability and consistency of the detection results.

[0003] Therefore, a donkey-hide gelatin quality detection method and system based on image recognition are proposed. Summary of the Invention

[0004] The purpose of the present invention is to provide a donkey-hide gelatin quality detection method and system based on image recognition. First, a time-series image data set at different stages of the donkey-hide gelatin production process is obtained through a multispectral camera, including boiling, filtering, concentration, and cooling and forming. Secondly, a donkey-hide gelatin image abnormal area recognition model is constructed to process the obtained time-series image data and identify the abnormal areas in the images at each stage. Then, based on the abnormal areas detected in the images at each stage, a donkey-hide gelatin image abnormal area detection model is constructed to compare and analyze the changes in the abnormal areas of the images, obtain the impurities abnormality of the donkey-hide gelatin image, and thus obtain the second area abnormal change coefficient. At the same time, through multi-parameter fusion technology, the abnormal change coefficients at different stages are integrated to obtain the donkey-hide gelatin abnormal coefficient. Finally, according to the donkey-hide gelatin abnormal coefficient, the donkey-hide gelatin is classified into first grade, second grade, and third grade, providing a standardized quality assessment mechanism.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A donkey-hide gelatin quality detection method based on image recognition, comprising:

[0007] Identifying the production process of donkey-hide gelatin, where the production process includes boiling, filtering, concentration, and cooling and forming;

[0008] Obtaining a time-series image data set in the production process, where the time-series image data set includes a first image, a second image, a third image, a fourth image, and a fifth image;

[0009] Construct an identification model for abnormal areas of donkey-hide gelatin images to identify a time-series image dataset, and obtain a first image abnormal area, a second image abnormal area, a third image abnormal area, a fourth image abnormal area, and a fifth image abnormal area;

[0010] Construct a detection model for abnormal areas of donkey-hide gelatin images to analyze the first image abnormal area, the second image abnormal area, the third image abnormal area, the fourth image abnormal area, and the fifth image abnormal area, and obtain a first regional abnormal change coefficient, a second regional abnormal change coefficient, a third regional abnormal change coefficient, and a fourth regional abnormal change coefficient;

[0011] Segment the donkey-hide gelatin after cooling and forming to obtain a first piece of donkey-hide gelatin;

[0012] According to the first regional abnormal change coefficient, the second regional abnormal change coefficient, the third regional abnormal change coefficient, and the fourth regional abnormal change coefficient, obtain a donkey-hide gelatin abnormal coefficient through multi-parameter fusion;

[0013] Classify the first piece of donkey-hide gelatin according to the donkey-hide gelatin abnormal coefficient to obtain first-class donkey-hide gelatin, second-class donkey-hide gelatin, and third-class donkey-hide gelatin.

[0014] Preferably, the images in the time-series image dataset are acquired by a multispectral camera;

[0015] The first image is an image acquired before boiling, the second image includes an image between after boiling and before filtration, the third image includes an image between after filtration and before concentration, the fourth image includes an image between after concentration and before cooling and forming, and the fifth image is an image after cooling and forming.

[0016] Preferably, the identification model for abnormal areas of donkey-hide gelatin images includes an image defect module, an image grayscale defect module, an image thermal defect analysis module, an image defect area fusion module, and an abnormal area output module;

[0017] The image defect module identifies the surface of the image to obtain a first image defect;

[0018] The image grayscale defect module grayscales the image to obtain a grayscale image; and obtains a second image defect based on the grayscale image;

[0019] The image thermal defect analysis module obtains a thermal map of the image and obtains a third image defect based on the temperature uniformity;

[0020] The image defect area fusion module determines the abnormal areas in the first image, the second image, the third image, the fourth image, and the fifth image according to the first image defect, the second image defect, and the third image defect;

[0021] The abnormal area output module marks and outputs the abnormal areas in the first image, the second image, the third image, the fourth image, and the fifth image.

[0022] Preferably, the image defect module is used to identify impurities in the donkey-hide gelatin production process; the impurities are hair, grease, incompletely dissolved colloid, and foam scum;

[0023] The image grayscale module is used to identify the color uniformity in the donkey-hide gelatin production process;

[0024] The image thermal defect analysis module is used for the temperature abnormal areas in the third image, the fourth image, and the fifth image before and after high-temperature concentration.

[0025] Preferably, the donkey-hide gelatin image abnormal area detection model includes a first donkey-hide gelatin impurity abnormal detection module, a second donkey-hide gelatin impurity abnormal detection module, a third donkey-hide gelatin impurity abnormal detection module, and a fourth donkey-hide gelatin impurity abnormal detection module;

[0026] The first donkey-hide gelatin impurity abnormal detection module analyzes the abnormal areas of the first image and the second image to obtain the first donkey-hide gelatin impurity abnormality, and the first donkey-hide gelatin impurity abnormality includes changes in the number of abnormal impurities, the diffusion range, and regional transformation; the first regional abnormal change coefficient is obtained based on the first donkey-hide gelatin impurity abnormality.

[0027] Preferably, the second donkey-hide gelatin impurity abnormal detection module analyzes the abnormal areas of the second image and the third image to obtain the second donkey-hide gelatin impurity abnormality, and the second donkey-hide gelatin impurity abnormality includes changes in the number of abnormal impurities, the diffusion range, and regional transformation; the second regional abnormal change coefficient is obtained based on the second donkey-hide gelatin impurity abnormality;

[0028] The third donkey-hide gelatin impurity abnormal detection module analyzes the abnormal areas of the third image and the fourth image to obtain the third donkey-hide gelatin impurity abnormality, and the third donkey-hide gelatin impurity abnormality includes temperature abnormality, color abnormality, changes in the number of abnormal impurities, the diffusion range, and regional transformation; the third regional abnormal change coefficient is obtained based on the third donkey-hide gelatin impurity abnormality;

[0029] The fourth donkey-hide gelatin impurity abnormal detection module analyzes the abnormal areas of the fourth image and the fifth image to obtain the fourth donkey-hide gelatin impurity abnormality, and the fourth donkey-hide gelatin impurity abnormality includes temperature abnormality, color abnormality, changes in the number of abnormal impurities, the diffusion range, and regional transformation; the fourth regional abnormal change coefficient is obtained based on the fourth donkey-hide gelatin impurity abnormality.

[0030] Preferably, the Ejiao anomaly coefficient includes the first region anomaly change coefficient, the second region anomaly change coefficient, the third region anomaly change coefficient, and the fourth region anomaly change coefficient, and the Ejiao anomaly coefficient is obtained through multi-parameter fusion; the specific calculation formula of the Ejiao anomaly coefficient is:

[0031] E = ω 1 E 1 + ω 2 E 2 + ω 3 E 3 + ω 4 E 4 ;

[0032] Among them, ω 1 is the weight of the first region anomaly change coefficient, E 1 is the first region anomaly change coefficient, ω 2 is the weight of the second region anomaly change coefficient, E 2 is the second region anomaly change coefficient, ω 3 is the weight of the third region anomaly change coefficient, E 3 is the third region anomaly change coefficient, ω 4 is the weight of the fourth region anomaly change coefficient, E 4 is the fourth region anomaly change coefficient;

[0033] According to the Ejiao anomaly coefficient, the first Ejiao is classified into grades, and the grade classification includes:

[0034] Set the abnormal grade threshold set TH = {S 1 , S 2}, when the Ejiao anomaly coefficient is less than S 1 , the corresponding first Ejiao grade is the first-class Ejiao; when the Ejiao anomaly coefficient is in the interval [S 1 , S 2 , the corresponding first Ejiao grade is the second-class Ejiao; when the Ejiao anomaly coefficient is greater than S 2 , the corresponding first Ejiao grade is the third-class Ejiao; the lower the first Ejiao grade, the better the quality of the first Ejiao.

[0035] An Ejiao quality detection system based on image recognition includes:

[0036] Identify the production process of Ejiao, and the production process includes boiling, filtering, concentration, and cooling and forming;

[0037] An Ejiao image acquisition unit for acquiring a time-series image dataset in the production process, and the time-series image dataset includes a first image, a second image, a third image, a fourth image, and a fifth image;

[0038] The first donkey-hide gelatin model construction unit is used to construct a donkey-hide gelatin image abnormal area recognition model to recognize a time-series image data set, and obtain a first image abnormal area, a second image abnormal area, a third image abnormal area, a fourth image abnormal area, and a fifth image abnormal area;

[0039] The second donkey-hide gelatin model construction unit is used to construct a donkey-hide gelatin image abnormal area detection model to analyze the first image abnormal area, the second image abnormal area, the third image abnormal area, the fourth image abnormal area, and the fifth image abnormal area, and obtain a first area abnormal change coefficient, a second area abnormal change coefficient, a third area abnormal change coefficient, and a fourth area abnormal change coefficient;

[0040] The donkey-hide gelatin abnormality acquisition unit is used to obtain a donkey-hide gelatin abnormality coefficient through multi-parameter fusion according to the first area abnormal change coefficient, the second area abnormal change coefficient, the third area abnormal change coefficient, and the fourth area abnormal change coefficient;

[0041] The donkey-hide gelatin grade acquisition unit is used to classify the first donkey-hide gelatin according to the donkey-hide gelatin abnormality coefficient, and obtain first-class donkey-hide gelatin, second-class donkey-hide gelatin, and third-class donkey-hide gelatin.

[0042] Preferably, the donkey-hide gelatin image abnormal area recognition model includes an image defect module, an image grayscale defect module, an image thermal defect analysis module, an image defect area fusion module, and an abnormal area output module.

[0043] Preferably, the first donkey-hide gelatin is classified according to the donkey-hide gelatin abnormality coefficient, and the classification includes:

[0044] Set an abnormal grade threshold set TH = {S 1 , S 2}, when the donkey-hide gelatin abnormality coefficient is less than S 1 , the corresponding first donkey-hide gelatin grade is first-class donkey-hide gelatin; when the donkey-hide gelatin abnormality coefficient is in the interval [S 1 , S 2 , the corresponding first donkey-hide gelatin grade is second-class donkey-hide gelatin; when the donkey-hide gelatin abnormality coefficient is greater than S 2 , the corresponding first donkey-hide gelatin grade is third-class donkey-hide gelatin; the lower the first donkey-hide gelatin grade, the better the quality of the first donkey-hide gelatin.

[0045] Compared with the prior art, the beneficial effects of the present invention are:

[0046] 1. The present invention collects and analyzes images at different stages in the production process of donkey-hide gelatin through image recognition technology based on a multispectral camera. By constructing a recognition model for abnormal regions in donkey-hide gelatin images, through the synergistic effects of the image defect module, grayscale defect module, and thermal defect analysis module, it enables comprehensive and accurate recognition of surface defects, color uniformity, and temperature anomalies during the high-temperature concentration process of donkey-hide gelatin. Compared with traditional manual detection methods, image recognition technology not only improves the detection efficiency but also reduces human errors, and can output quality detection results of donkey-hide gelatin in real time and efficiently, ensuring the consistency and reliability of product quality.

[0047] 2. The present invention constructs a detection model for abnormal regions in donkey-hide gelatin images. Based on the multi-stage detection mechanism of sequential images, it deeply analyzes subtle defects in the images, including means such as grayscale processing and heat map analysis, to ensure temperature changes, the number of impurities, diffusion range, and regional transformation indicators. This method not only enhances the comprehensiveness and sensitivity of detection but also reduces human intervention through the automated processing of image recognition technology, improving the detection efficiency and accuracy.

[0048] 3. The present invention realizes the automation and precision of donkey-hide gelatin quality detection through image recognition technology. First, it obtains sequential image data during the production process of donkey-hide gelatin through a multispectral camera, and combines the recognition and detection model for abnormal regions in the images to deeply analyze abnormal parameters such as impurities, temperature, and color at each stage, and accurately identifies defects in donkey-hide gelatin in real time. The abnormal coefficient of donkey-hide gelatin is calculated through multi-parameter fusion, and the donkey-hide gelatin is classified according to this coefficient. This method can effectively distinguish the quality level of donkey-hide gelatin, thereby ensuring the stability and consistency of product quality, reducing human errors, and improving production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a schematic flowchart of a donkey-hide gelatin quality detection method based on image recognition provided by an embodiment of the present invention;

[0050] Figure 2 It is a schematic structural diagram of a donkey-hide gelatin quality detection system based on image recognition provided by an embodiment of the present invention;

[0051] Figure 3 It is a schematic structural diagram of a recognition model for abnormal regions in donkey-hide gelatin images provided by an embodiment of the present invention;

[0052] Figure 4 It is a schematic structural diagram of a detection model for abnormal regions in donkey-hide gelatin images provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0053] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0054] The present invention provides a donkey-hide gelatin quality detection method based on image recognition. This method is applied to a donkey-hide gelatin quality detection system based on image recognition. For the specific method and the flowchart of the system, refer to Figure 1 and Figure 2 .

[0055] Embodiment 1

[0056] The main raw material of donkey-hide gelatin is donkey skin. Before making donkey-hide gelatin, the donkey skin needs to be cleaned and depilated to remove surface impurities and hair. The donkey skin is soaked in water and repeatedly rubbed to remove dirt, fat, and hair on the leather. The processed donkey skin is cut into small pieces for boiling. Before boiling, an image is obtained as the first image; the gelatin component in the donkey skin is boiled out to form a liquid rich in collagen, and at this time, an image after boiling is obtained; impurities and residues will be generated after boiling, and these impurities are removed by filtration. Images are obtained before and after filtration respectively. As an implementation manner of the present invention, refer to Figure 1 S10 in. S10 is applied to the donkey-hide gelatin image acquisition unit of a donkey-hide gelatin quality detection system based on image recognition. The donkey-hide gelatin image acquisition unit is used to acquire a time-series image dataset in the production process. The time-series image dataset includes a first image, a second image, a third image, a fourth image, and a fifth image. The images in the time-series image dataset are acquired by a multispectral camera. The data of the production line of donkey-hide gelatin A are collected in this embodiment.

[0057] Furthermore, the first image is the image obtained before boiling, the second image includes the image between after boiling and before filtration, the third image includes the image between after filtration and before concentration, the fourth image includes the image between after concentration and before cooling and forming, and the fifth image is the image after cooling and forming.

[0058] In this embodiment, time-series images are acquired by a multispectral camera during the production process. This method captures the changes of donkey-hide gelatin at different stages, providing multi-dimensional data for subsequent operations. These data provide accurate image recognition and analysis for the quality and defect abnormal conditions corresponding to each stage, helping to accurately locate problems in the production process, so as to take corrective measures in a timely manner and optimize the production process.

[0059] Refer to Figure 1In S20, S20 is applied to the first donkey-hide gelatin model construction unit of a donkey-hide gelatin quality detection system based on image recognition. The first donkey-hide gelatin model construction unit is used to construct a donkey-hide gelatin image abnormal area recognition model to recognize a time-series image data set, and obtain a first image abnormal area, a second image abnormal area, a third image abnormal area, a fourth image abnormal area and a fifth image abnormal area;

[0060] Furthermore, the donkey-hide gelatin image abnormal area recognition model includes an image defect module, an image grayscale defect module, an image thermal defect analysis module, an image defect area fusion module and an abnormal area output module. Refer to Figure 3 ;

[0061] The image defect module obtains a first image defect by recognizing the image surface;

[0062] The image grayscale defect module obtains a grayscale image by performing grayscale processing on the image; obtains a second image defect based on the grayscale image; the image thermal defect analysis module obtains a third image defect by obtaining a thermal map of the image and based on the temperature uniformity; the image defect area fusion module determines the abnormal areas in the first image, the second image, the third image, the fourth image and the fifth image according to the first image defect, the second image defect and the third image defect; the abnormal area output module marks and outputs the abnormal areas in the first image, the second image, the third image, the fourth image and the fifth image.

[0063] In this embodiment, by constructing a donkey-hide gelatin image abnormal area recognition model to analyze each image in the time-series image data set one by one, the abnormal stages in the donkey-hide gelatin production process are obtained. At the same time, image recognition technology is also used to locate the defects in the donkey-hide gelatin, such as impurities, bubbles and poor molding. This model improves the detection accuracy for donkey-hide gelatin quality monitoring.

[0064] Furthermore, the image defect module is used to identify the impurities in the donkey-hide gelatin production process; the impurities are hair, grease, incompletely dissolved colloid and foam scum;

[0065] The image grayscale module is used to identify the color uniformity in the donkey-hide gelatin production process;

[0066] The image thermal defect analysis module is used for the temperature abnormal areas of the third image, the fourth image and the fifth image before and after high-temperature concentration.

[0067] In this embodiment, through the recognition of abnormal image regions, in-depth analysis is carried out on the defective regions in the image; the analysis methods include means such as grayscale processing and heat map analysis, so as to identify the number of impurities, the diffusion range, and the regional transformation in the image. This method not only enhances the comprehensiveness of detection, but also reduces human intervention through the automated processing of image recognition technology, greatly improving the detection efficiency and accuracy.

[0068] Referring to Figure 1 S30 in, S30 is applied to the second donkey-hide gelatin model construction unit of a donkey-hide gelatin quality detection system based on image recognition. The second donkey-hide gelatin model construction unit is used to construct a donkey-hide gelatin image abnormal region detection model to analyze the first image abnormal region, the second image abnormal region, the third image abnormal region, the fourth image abnormal region, and the fifth image abnormal region, and obtain the first region abnormal change coefficient, the second region abnormal change coefficient, the third region abnormal change coefficient, and the fourth region abnormal change coefficient;

[0069] Furthermore, the donkey-hide gelatin image abnormal region detection model includes a first donkey-hide gelatin impurity abnormal detection module, a second donkey-hide gelatin impurity abnormal detection module, a third donkey-hide gelatin impurity abnormal detection module, and a fourth donkey-hide gelatin impurity abnormal detection module. Refer to Figure 4 ;

[0070] The first donkey-hide gelatin impurity abnormal detection module is used to analyze the first image abnormal region and the second image abnormal region to obtain the first donkey-hide gelatin impurity abnormal, and the first donkey-hide gelatin impurity abnormal includes abnormal impurity quantity change, diffusion range, and regional transformation; the first region abnormal change coefficient is obtained based on the first donkey-hide gelatin impurity abnormal.

[0071] The second donkey-hide gelatin impurity abnormal detection module is used to analyze the second image abnormal region and the third image abnormal region to obtain the second donkey-hide gelatin impurity abnormal, and the second donkey-hide gelatin impurity abnormal includes abnormal impurity quantity change, diffusion range, and regional transformation; the second region abnormal change coefficient is obtained based on the second donkey-hide gelatin impurity abnormal;

[0072] The third donkey-hide gelatin impurity abnormal detection module is used to analyze the third image abnormal region and the fourth image abnormal region to obtain the third donkey-hide gelatin impurity abnormal, and the third donkey-hide gelatin impurity abnormal includes temperature abnormality, color abnormality, abnormal impurity quantity change, diffusion range, and regional transformation; the third region abnormal change coefficient is obtained based on the third donkey-hide gelatin impurity abnormal;

[0073] The fourth donkey-hide gelatin impurity anomaly detection module is used to analyze the fourth image anomaly region and the fifth image anomaly region to obtain a fourth donkey-hide gelatin impurity anomaly, where the fourth donkey-hide gelatin impurity anomaly includes temperature anomaly, color anomaly, change in the number of abnormal impurities, diffusion range, and regional transformation; a fourth regional anomaly change coefficient is obtained based on the fourth donkey-hide gelatin impurity anomaly.

[0074] When temperature anomaly, color anomaly, change in the number of abnormal impurities, diffusion range, and regional transformation change, the resulting effects are positively or negatively correlated. For example, when the change in the number of abnormal impurities in the second donkey-hide gelatin impurity anomaly detection module is less than that in the first donkey-hide gelatin impurity anomaly detection module, the change rate of the number of abnormal impurities is negative; the regional anomaly change coefficient is obtained by weighted average calculation of each change rate, specifically referring to Table 1;

[0075] Table 1 Change Rates of Donkey-Hide Gelatin Impurity Anomaly Detection

[0076]

[0077] Among them, the images detected by the second donkey-hide gelatin impurity anomaly detection module are the images before and after filtration. Therefore, the number of impurities and the diffusion range are reduced compared to the number of impurities and the diffusion range in the first donkey-hide gelatin impurity anomaly detection module; the temperature difference in the third and fourth donkey-hide gelatin impurity anomaly detection modules is due to the fact that the fourth image is an image after high-temperature concentration, and the temperature of donkey-hide gelatin needs to reach 120°C during high-temperature concentration; the color anomaly difference is caused by uneven heating at different positions during high-temperature concentration and cooling of donkey-hide gelatin, and the color difference calculation is performed by the representation method of color space.

[0078] In this embodiment, the image anomaly region detection model provides a scientific basis for the comprehensive evaluation of donkey-hide gelatin quality by analyzing the anomaly regions in multiple images. The model calculates the anomaly change coefficients of each image anomaly region based on the change situations of the anomaly regions in different images, and identifies temperature anomalies, uneven colors, and impurity diffusion during the production process of donkey-hide gelatin by analyzing the changes in the anomaly regions between images, thereby effectively capturing the minute quality changes during the production process of donkey-hide gelatin and providing detailed data support for subsequent quality optimization.

[0079] Referring to Figure 1 In S40 of, S40 is applied to the donkey-hide gelatin anomaly acquisition unit of an image recognition-based donkey-hide gelatin quality detection system. The donkey-hide gelatin anomaly acquisition unit is used to obtain a donkey-hide gelatin anomaly coefficient through multi-parameter fusion based on the first regional anomaly change coefficient, the second regional anomaly change coefficient, the third regional anomaly change coefficient, and the fourth regional anomaly change coefficient;

[0080] The Ejiao anomaly acquisition unit fuses all the anomaly coefficients to generate an Ejiao anomaly coefficient. First, the anomaly coefficients of color, uniformity, impurities, and bubbles are integrated into a preliminary anomaly assessment result, and then through further fusion with the anomaly coefficients generated in the first stage and the current stage, the Ejiao anomaly coefficient is obtained.

[0081] This fusion process combines the anomaly detection results of multiple stages by assigning different weights to form a global quality evaluation index. The Ejiao anomaly coefficient is ultimately used to determine whether the Ejiao product meets the production standards, ensuring that the produced Ejiao reaches consistency and high standards in terms of appearance, internal structure, and quality. The multi-step processing and multi-dimensional analysis method of this model provide strong data support and guarantee for quality control in Ejiao production.

[0082] The Ejiao anomaly coefficient includes the first region anomaly change coefficient, the second region anomaly change coefficient, the third region anomaly change coefficient, and the fourth region anomaly change coefficient, and the Ejiao anomaly coefficient is obtained through multi-parameter fusion. The specific calculation formula for the Ejiao anomaly coefficient is:

[0083] E = ω 1 E 1 + ω 2 E 2 + ω 3 E 3 + ω 4 E 4 ;

[0084] Where, ω 1 is the weight of the first region anomaly change coefficient, E 1 is the first region anomaly change coefficient, ω 2 is the weight of the second region anomaly change coefficient, E 2 is the second region anomaly change coefficient, ω 3 is the weight of the third region anomaly change coefficient, E 3 is the third region anomaly change coefficient, ω 4 is the weight of the fourth region anomaly change coefficient, E 4 is the fourth region anomaly change coefficient.

[0085] In this embodiment, the Ejiao anomaly coefficient is calculated through the weighted combination of multiple features. Since the Ejiao anomaly coefficient is the fusion of multiple stages and multi-dimensional features, it not only provides a comprehensive analysis for Ejiao quality monitoring but also combines the key quality control points in the entire production process to ensure the qualification and consistency of the final product.

[0086] Refer to Figure 1In S50, S50 is applied to the Ejiao grade acquisition unit of an Ejiao quality detection system based on image recognition. The Ejiao grade acquisition unit is used to classify the first Ejiao according to the Ejiao abnormality coefficient, and obtain first-grade Ejiao, second-grade Ejiao, and third-grade Ejiao.

[0087] Further, the first Ejiao is classified according to the Ejiao abnormality coefficient, and the classification includes:

[0088] Set the abnormal grade threshold set TH = {S 1 , S 2}. When the Ejiao abnormality coefficient is less than S 1 , the corresponding first Ejiao grade is first-grade Ejiao, which is the highest quality grade, indicating that the product has the least abnormalities in terms of color, uniformity, impurities, and bubbles, and the overall quality is close to the ideal state. When the Ejiao abnormality coefficient is in the interval [S 1 , S 2 , the corresponding first Ejiao grade is second-grade Ejiao. Although there are certain abnormalities in this type of Ejiao, it still meets most quality standards and is suitable for general market demand. When the Ejiao abnormality coefficient is greater than S 2 , the corresponding first Ejiao grade is third-grade Ejiao, indicating that the product has relatively significant quality problems, such as more impurities, obvious bubbles, or uneven surface, and is suitable for price-sensitive markets. The lower the first Ejiao grade, the better the quality of the first Ejiao. Among them, S 1 = 0.1, S 2 = 0.3. For the specific Ejiao grade classification, refer to Table 2;

[0089] Table 2 Ejiao grade classification

[0090]

[0091] Among them, the weight ratio is obtained by training with historical Ejiao production data, and ω 4 > ω 3 > ω 2 > ω 1 .

[0092] In this embodiment, the grade classification mechanism of the Ejiao abnormality coefficient realizes the grade classification of Ejiao with different quality levels. At the same time, the application of the Ejiao abnormality coefficient can not only control the production quality, but also provide suitable products for different customer groups, thereby enhancing the market competitiveness and production efficiency of Ejiao products.

[0093] The method for detecting the quality of Ejiao based on image recognition technology of the present invention has significant practical value in improving the transparency of Ejiao production and quality control. First, the present invention collects sequential images of each key link in the Ejiao production process through a multispectral camera, and combines image defect recognition, grayscale processing, and heat map analysis to identify potential quality abnormalities in the Ejiao production process, where the quality abnormalities include impurities, uneven color, and temperature fluctuations, etc. Secondly, this method further analyzes the abnormal areas between the detected images by constructing an abnormal area detection model to obtain the changes in quality abnormalities between the images. Finally, based on the multi-parameter fusion technology, the abnormal coefficient of Ejiao is obtained, so as to realize the quality evaluation and grading of the quality of Ejiao. Compared with traditional quality detection means, the Ejiao detection method based on image recognition of the present invention not only greatly improves the detection accuracy and efficiency of Ejiao, but also provides a scientific basis for the quality determination of Ejiao.

[0094] Embodiment 2

[0095] In another embodiment of the present invention, a quality detection system for Ejiao based on image recognition is also introduced. For details, refer to Figure 1 and Figure 2 ;

[0096] Identify the production process of Ejiao, and the production process includes boiling, filtering, concentration, and cooling and forming;

[0097] The Ejiao image acquisition unit is used to acquire a sequential image data set in the production process, and the sequential image data set includes a first image, a second image, a third image, a fourth image, and a fifth image;

[0098] The first Ejiao model construction unit is used to construct an Ejiao image abnormal area recognition model to identify the sequential image data set, and obtain a first image abnormal area, a second image abnormal area, a third image abnormal area, a fourth image abnormal area, and a fifth image abnormal area;

[0099] The second Ejiao model construction unit is used to construct an Ejiao image abnormal area detection model to analyze the first image abnormal area, the second image abnormal area, the third image abnormal area, the fourth image abnormal area, and the fifth image abnormal area, and obtain a first area abnormal change coefficient, a second area abnormal change coefficient, a third area abnormal change coefficient, and a fourth area abnormal change coefficient;

[0100] The Ejiao abnormality acquisition unit is used to obtain the Ejiao abnormal coefficient through multi-parameter fusion according to the first area abnormal change coefficient, the second area abnormal change coefficient, the third area abnormal change coefficient, and the fourth area abnormal change coefficient;

[0101] The Ejiao grade acquisition unit is used to classify the first Ejiao according to the Ejiao anomaly coefficient, and obtain first-grade Ejiao, second-grade Ejiao, and third-grade Ejiao.

[0102] Classify the first Ejiao according to the Ejiao anomaly coefficient, and the classification includes:

[0103] The Ejiao anomaly coefficient includes the first region anomaly change coefficient, the second region anomaly change coefficient, the third region anomaly change coefficient, and the fourth region anomaly change coefficient, and the Ejiao anomaly coefficient is obtained through multi-parameter fusion; the specific calculation formula of the Ejiao anomaly coefficient is:

[0104] E = ω 1 E 1 + ω 2 E 2 + ω 3 E 3 + ω 4 E 4 ;

[0105] Among them, ω 1 is the weight of the first region anomaly change coefficient, E 1 is the first region anomaly change coefficient, ω 2 is the weight of the second region anomaly change coefficient, E 2 is the second region anomaly change coefficient, ω 3 is the weight of the third region anomaly change coefficient, E 3 is the third region anomaly change coefficient, ω 4 is the weight of the fourth region anomaly change coefficient, E 4 is the fourth region anomaly change coefficient;

[0106] Set the anomaly level threshold set TH = {S 1 , S 2}, when the Ejiao anomaly coefficient is less than S 1 , the corresponding first Ejiao grade is first-grade Ejiao; when the Ejiao anomaly coefficient is in the interval [S 1 , S 2 )), the corresponding first Ejiao grade is second-grade Ejiao; when the Ejiao anomaly coefficient is greater than S 2 , the corresponding first Ejiao grade is third-grade Ejiao; the lower the first Ejiao grade, the better the quality of the first Ejiao.

[0107] The present invention provides a donkey-hide gelatin quality detection system based on image recognition, thereby enhancing the quality control ability in the production process of donkey-hide gelatin. First, the present invention comprehensively monitors key links such as boiling, filtering, concentrating, and cooling and forming by acquiring a time-series image dataset during the production process of donkey-hide gelatin. Secondly, an image abnormal area recognition model is used to identify and analyze features such as impurities, color, and temperature in the images. This model not only improves the accuracy of donkey-hide gelatin detection but also provides a data basis for the quality evaluation of donkey-hide gelatin. By calculating the abnormal change coefficient of donkey-hide gelatin and grading the quality of donkey-hide gelatin according to the industry standard threshold, the subjectivity and error of manual detection are significantly reduced, ensuring the quality consistency of donkey-hide gelatin of different grades.

[0108] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for detecting donkey-hide gelatin quality based on image recognition, characterized in that: include: Identify the production process of donkey-hide gelatin, which includes boiling, filtering, concentrating and cooling and molding; Acquire a time-series image data set in a production process, the time-series image data set comprising a first image, a second image, a third image, a fourth image and a fifth image; the first image is an image acquired before brewing, the second image comprises an image after brewing and before filtering, the third image comprises an image after filtering and before concentration, the fourth image comprises an image after concentration and before cooling and forming, and the fifth image is an image after cooling and forming; Constructing an abnormal region recognition model for donkey-hide gelatin images to recognize the time series image data set, and obtaining a first image abnormal region, a second image abnormal region, a third image abnormal region, a fourth image abnormal region, and a fifth image abnormal region; Constructing an abnormal region detection model for donkey-hide gelatin images to analyze the first image abnormal region, the second image abnormal region, the third image abnormal region, the fourth image abnormal region, and the fifth image abnormal region to obtain the first region abnormal variation coefficient, the second region abnormal variation coefficient, the third region abnormal variation coefficient, and the fourth region abnormal variation coefficient; The abnormal area detection model of donkey-hide gelatin image includes a first donkey-hide gelatin impurity abnormality detection module, a second donkey-hide gelatin impurity abnormality detection module, a third donkey-hide gelatin impurity abnormality detection module and a fourth donkey-hide gelatin impurity abnormality detection module; The first donkey-hide gelatin impurity anomaly detection module is used to analyze the first image abnormal area and the second image abnormal area to obtain the first donkey-hide gelatin impurity anomaly, wherein the first donkey-hide gelatin impurity anomaly includes the change in the number of abnormal impurities, the diffusion range and the regional transformation; and obtain the first regional abnormal change coefficient according to the first donkey-hide gelatin impurity anomaly; The second donkey-hide gelatin impurity abnormality detection module is used to analyze the abnormal area of ​​the second image and the abnormal area of ​​the third image to obtain the second donkey-hide gelatin impurity abnormality, wherein the second donkey-hide gelatin impurity abnormality includes the change in the number of abnormal impurities, the diffusion range and the regional transformation; and obtain the second regional abnormality change coefficient according to the second donkey-hide gelatin impurity abnormality; The third donkey-hide gelatin impurity anomaly detection module is used to analyze the third image abnormal area and the fourth image abnormal area to obtain the third donkey-hide gelatin impurity anomaly, wherein the third donkey-hide gelatin impurity anomaly includes temperature anomaly, color anomaly, change in the number of abnormal impurities, diffusion range and regional change; and obtain the third regional abnormal change coefficient according to the third donkey-hide gelatin impurity anomaly; The fourth donkey-hide gelatin impurity anomaly detection module is used to analyze the fourth image abnormal area and the fifth image abnormal area to obtain the fourth donkey-hide gelatin impurity anomaly, wherein the fourth donkey-hide gelatin impurity anomaly includes temperature anomaly, color anomaly, abnormal impurity quantity change, diffusion range and regional change; and obtain the fourth regional abnormal change coefficient according to the fourth donkey-hide gelatin impurity anomaly; The cooled and formed donkey-hide gelatin is divided to obtain the first donkey-hide gelatin; Obtaining the donkey-hide gelatin abnormality coefficient through multi-parameter fusion according to the abnormal variation coefficient of the first region, the abnormal variation coefficient of the second region, the abnormal variation coefficient of the third region, and the abnormal variation coefficient of the fourth region; The first donkey-hide gelatin is graded according to the donkey-hide gelatin abnormality coefficient to obtain first-grade donkey-hide gelatin, second-grade donkey-hide gelatin and third-grade donkey-hide gelatin.

2. The method for detecting donkey-hide gelatin quality based on image recognition according to claim 1, characterized in that: The abnormal area recognition model of donkey-hide gelatin image includes an image defect module, an image grayscale defect module, an image thermal defect analysis module, an image defect area fusion module and an abnormal area output module; The image defect module obtains a first image defect by identifying the image surface; The image grayscale defect module obtains a grayscale image by grayscale processing the image; and obtains a second image defect according to the grayscale image; The image thermal defect analysis module obtains a third image defect according to the temperature uniformity by acquiring a thermal map of the image; The image defect area fusion module determines abnormal areas in the first image, the second image, the third image, the fourth image, and the fifth image according to the first image defect, the second image defect, and the third image defect; The abnormal region output module marks and outputs abnormal regions in the first image, the second image, the third image, the fourth image, and the fifth image.

3. The method for detecting donkey-hide gelatin quality based on image recognition according to claim 2, characterized in that: The image defect module is used to identify impurities in the donkey-hide gelatin production process; the impurities are hair, grease, incompletely dissolved colloids and foam scum; The image grayscale defect module is used to identify the uniformity of color in the donkey-hide gelatin production process; The image thermal defect analysis module is used for temperature abnormality areas of the third image, the fourth image and the fifth image before and after high-temperature concentration.

4. The method for detecting donkey-hide gelatin quality based on image recognition according to claim 1, characterized in that: The donkey-hide gelatin abnormal coefficient includes the abnormal variation coefficient of the first region, the abnormal variation coefficient of the second region, the abnormal variation coefficient of the third region and the abnormal variation coefficient of the fourth region, and the donkey-hide gelatin abnormal coefficient is obtained by multi-parameter fusion; the specific calculation formula of the donkey-hide gelatin abnormal coefficient is: ； in, is the first regional anomaly variation coefficient weight, is the anomaly variation coefficient of the first region, is the weight of the anomaly variation coefficient of the second region, is the anomaly variation coefficient of the second region, is the third area abnormal variation coefficient weight, is the anomaly variation coefficient of the third region, is the fourth region abnormal variation coefficient weight, is the anomaly variation coefficient of the fourth region; The first donkey-hide gelatin is graded according to the donkey-hide gelatin abnormality coefficient, and the grade classification includes: Set anomaly level threshold set , when the abnormal coefficient of donkey-hide gelatin is less than When the first donkey-hide gelatin grade is first-grade donkey-hide gelatin; when the donkey-hide gelatin abnormality coefficient is in the interval When the first donkey-hide gelatin grade is the second-grade donkey-hide gelatin, the abnormal coefficient of the donkey-hide gelatin is greater than The corresponding first donkey-hide gelatin grade is third-grade donkey-hide gelatin; the lower the first donkey-hide gelatin grade, the better the quality of the first donkey-hide gelatin.

5. A donkey-hide gelatin quality detection system based on image recognition, characterized in that: include: Identify the production process of donkey-hide gelatin, which includes boiling, filtering, concentrating and cooling and molding; The donkey-hide gelatin image acquisition unit is used to acquire a time-series image data set in the production process, wherein the time-series image data set includes a first image, a second image, a third image, a fourth image, and a fifth image; the first image is an image acquired before boiling, the second image includes an image between boiling and before filtering, the third image includes an image between filtering and before concentration, the fourth image includes an image between concentration and before cooling and molding, and the fifth image is an image after cooling and molding; A first donkey-hide gelatin model construction unit is used to construct a donkey-hide gelatin image abnormal region recognition model to recognize the time series image data set, and obtain a first image abnormal region, a second image abnormal region, a third image abnormal region, a fourth image abnormal region, and a fifth image abnormal region; The second donkey-hide gelatin model construction unit is used to construct a donkey-hide gelatin image abnormal region detection model to analyze the first image abnormal region, the second image abnormal region, the third image abnormal region, the fourth image abnormal region and the fifth image abnormal region to obtain the first region abnormal variation coefficient, the second region abnormal variation coefficient, the third region abnormal variation coefficient and the fourth region abnormal variation coefficient; The abnormal area detection model of donkey-hide gelatin image includes a first donkey-hide gelatin impurity abnormality detection module, a second donkey-hide gelatin impurity abnormality detection module, a third donkey-hide gelatin impurity abnormality detection module and a fourth donkey-hide gelatin impurity abnormality detection module; The first donkey-hide gelatin impurity anomaly detection module is used to analyze the first image abnormal area and the second image abnormal area to obtain the first donkey-hide gelatin impurity anomaly, wherein the first donkey-hide gelatin impurity anomaly includes the change in the number of abnormal impurities, the diffusion range and the regional transformation; and obtain the first regional abnormal change coefficient according to the first donkey-hide gelatin impurity anomaly; The second donkey-hide gelatin impurity abnormality detection module is used to analyze the abnormal area of ​​the second image and the abnormal area of ​​the third image to obtain the second donkey-hide gelatin impurity abnormality, wherein the second donkey-hide gelatin impurity abnormality includes the change in the number of abnormal impurities, the diffusion range and the regional transformation; and obtain the second regional abnormality change coefficient according to the second donkey-hide gelatin impurity abnormality; The third donkey-hide gelatin impurity anomaly detection module is used to analyze the third image abnormal area and the fourth image abnormal area to obtain the third donkey-hide gelatin impurity anomaly, wherein the third donkey-hide gelatin impurity anomaly includes temperature anomaly, color anomaly, change in the number of abnormal impurities, diffusion range and regional change; and obtain the third regional abnormal change coefficient according to the third donkey-hide gelatin impurity anomaly; The fourth donkey-hide gelatin impurity anomaly detection module is used to analyze the fourth image abnormal area and the fifth image abnormal area to obtain the fourth donkey-hide gelatin impurity anomaly, wherein the fourth donkey-hide gelatin impurity anomaly includes temperature anomaly, color anomaly, abnormal impurity quantity change, diffusion range and regional change; and obtain the fourth regional abnormal change coefficient according to the fourth donkey-hide gelatin impurity anomaly; The cooled and formed donkey-hide gelatin is divided to obtain the first donkey-hide gelatin; A donkey-hide gelatin abnormality acquisition unit, configured to obtain the donkey-hide gelatin abnormality coefficient through multi-parameter fusion according to the first region abnormality variation coefficient, the second region abnormality variation coefficient, the third region abnormality variation coefficient, and the fourth region abnormality variation coefficient; The donkey-hide gelatin grade acquisition unit is used to grade the first donkey-hide gelatin according to the donkey-hide gelatin abnormality coefficient to obtain first-grade donkey-hide gelatin, second-grade donkey-hide gelatin and third-grade donkey-hide gelatin.

6. The image recognition-based donkey-hide gelatin quality detection system according to claim 5, characterized in that: The abnormal area recognition model of donkey-hide gelatin image includes an image defect module, an image grayscale defect module, an image thermal defect analysis module, an image defect area fusion module and an abnormal area output module.

7. The image recognition-based donkey-hide gelatin quality detection system according to claim 5, characterized in that: The first donkey-hide gelatin is graded according to the donkey-hide gelatin abnormality coefficient, and the grade classification includes: Set anomaly level threshold set , when the abnormal coefficient of donkey-hide gelatin is less than When the first donkey-hide gelatin grade is first-grade donkey-hide gelatin; when the donkey-hide gelatin abnormality coefficient is in the interval When the first donkey-hide gelatin grade is the second-grade donkey-hide gelatin, the abnormal coefficient of the donkey-hide gelatin is greater than The corresponding first donkey-hide gelatin grade is third-grade donkey-hide gelatin; the lower the first donkey-hide gelatin grade, the better the quality of the first donkey-hide gelatin.

Images