Pork traceability management method and system based on big data

Abstract

The present application relates to the field of traceability, especially to a pork traceability management method and system based on big data, the present application obtains pork image samples and corresponding label information, classifies the samples based on the label information, and constructs confidence perspective sample features, determines the confidence perspective by analyzing the differences between the texture perspective features and the color perspective features, and then processes the collected production batch information, pork images and label information, in the processing stage, based on the extracted perspective features corresponding to the confidence perspective in the pork image, and matching with the confidence perspective sample features, to determine whether the image is abnormal, and tracing the production batch corresponding to the abnormal image, the present application combines the confidence perspective analysis and the abnormal identification model, improves the processing efficiency and accuracy of the traceability system.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of traceability, in particular to a pig meat traceability management method and system based on big data. BACKGROUND

[0002] With the rapid development of social economy and the improvement of consumer health awareness, food safety issues have attracted increasing attention. Especially in the pig breeding industry, how to ensure the safety and quality traceability of live pigs from birth to market has become an important issue in modern agricultural management. The live pig traceability management system has emerged as the times require and has become an important tool to ensure the quality and safety of pork. Especially in the context of smart agriculture, with the maturity of big data, Internet of Things, artificial intelligence, and blockchain technologies, the live pig traceability management system not only plays a huge role in traditional breeding, but also has great potential in promoting agricultural modernization and improving agricultural production efficiency.

[0003] Chinese patent publication No. CN118863928A discloses a pig meat information anomaly detection and traceability method based on the Internet of Things. A drug residue feature matrix of each pig is constructed. The drug residue difference matrix is obtained according to the drug residue feature matrix of each pig and other pigs, thereby obtaining abnormal pigs. A change factor set of each drug residue amount sequence of normal pigs is constructed. The change factor set of each abnormal pig is obtained. The abnormal drug residue amount sequence of each abnormal pig is obtained based on the change factor set of normal pigs. The peak difference of each peak value of each abnormal drug residue amount sequence of each abnormal pig is calculated to obtain each abnormal time point, and the abnormality of each pig is judged. The evaluation result of the abnormal pig is more comprehensive, the accuracy of abnormal pig screening is improved, and the accuracy of pork traceability positioning is improved.

[0004] However, the prior art still has the following problems,

[0005] Since the prior art usually relies on complex calculation models to analyze all pork images one by one, although this method has high detection accuracy, the model calculation itself consumes a large amount of computing resources and time, resulting in a decrease in the processing efficiency of the overall traceability system and an increase in operating costs. Especially in large-scale circulation scenarios, full-model reasoning on all batches will bring significant computational load, causing unnecessary resource waste and making it difficult to meet real-time or high-concurrency processing requirements, thereby reducing the operating efficiency of the traceability system. SUMMARY

[0006] To this end, the application provides a pork traceability management method and system based on big data, to overcome the problem in the prior art that all pork images are usually analyzed one by one by relying on a complex calculation model, although the detection accuracy is high, the model calculation itself consumes a large amount of computing resources and time, resulting in reduced processing efficiency of the overall traceability system and increased operating costs. In particular, in a large-scale circulation scenario, full-model reasoning on all batches will bring significant computing load, causing unnecessary waste of resources and making it difficult to meet real-time or high-concurrency processing requirements, thereby reducing the operation efficiency of the traceability system.

[0007] To achieve the above-mentioned purpose, in one aspect, the application provides a pork traceability management method based on big data, comprising:

[0008] Obtaining a plurality of pork image samples and corresponding label information, the label information including the origin, part and slaughter time interval corresponding to the pork image samples;

[0009] Classifying the pork image samples based on the label information to obtain a plurality of pork image sets, and performing confidence perspective analysis on each pork image set, including determining the confidence perspective for the label information based on the differences in texture perspective features and the differences in chrominance perspective features between the pork image samples, and simultaneously constructing confidence perspective sample features;

[0010] In response to collecting production batch information, pork images and label information, calling the confidence perspective associated with the label information, processing the pork images, including,

[0011] Extracting the perspective features corresponding to the confidence perspective in the pork images and matching the corresponding confidence perspective sample features, to determine whether the pork images are abnormal; or, analyzing the pork images through a pre-trained abnormality recognition model to determine whether the pork images are abnormal;

[0012] Verifying the abnormality recognition model for the non-abnormal pork images, and determining whether to update the confidence perspective sample features based on the verification result;

[0013] Determining that the production batch corresponding to the abnormal pork images needs to be traced;

[0014] The confidence perspective includes texture confidence perspective and chrominance confidence perspective, and the confidence perspective sample features include texture confidence perspective sample features and chrominance confidence perspective sample features.

[0015] Further, the process of classifying the pork image samples based on the label information to obtain a plurality of pork image sets, and performing confidence perspective analysis on each pork image set, comprises,

[0016] The pork image samples with the same label information are classified into the same category to obtain pork image sets;

[0017] The texture features and the chrominance features of the pork image samples in each pork image set are determined;

[0018] The lean meat regions of each pork image sample are extracted, and the texture interval variance of the lean meat regions between the pork image samples is determined as the difference of the texture perspective features;

[0019] The chrominance variance between the pork image samples is determined as the difference of the chrominance perspective features.

[0020] Further, the process of determining the confidence perspective for the label information based on the difference of the texture perspective features and the difference of the chrominance perspective features between the pork image samples comprises,

[0021] If the difference of the texture perspective features is less than a preset texture difference threshold, it is determined that the label information has a texture confidence perspective, and a texture confidence sample feature is constructed synchronously;

[0022] If the difference of the chrominance perspective features is less than a preset chrominance difference threshold, it is determined that the label information has a chrominance confidence perspective, and a chrominance confidence sample feature is constructed synchronously.

[0023] Further, the process of constructing the confidence perspective sample feature synchronously comprises,

[0024] The average fiber texture interval of the lean meat region in each pork image sample in each pork image set is extracted, and the average fiber texture interval is determined as the texture confidence perspective sample feature;

[0025] The average chrominance between the pork image samples in each pork image set is extracted, and the average chrominance is determined as the chrominance confidence perspective sample feature.

[0026] Further, the process of calling the confidence perspective associated with the label information to process the pork image comprises,

[0027] If the label information has a confidence perspective, the perspective feature corresponding to the confidence perspective in the pork image is extracted and matched with the confidence perspective sample feature corresponding to the confidence perspective to determine whether the pork image has an anomaly;

[0028] If the label information does not have a confidence perspective, the pork image is analyzed by a pre-trained anomaly recognition model to determine whether the pork image has an anomaly.

[0029] Further, the process of matching the view angle feature corresponding to the confidence view angle in the extracted pork image with the corresponding confidence view angle sample feature to determine whether the pork image is abnormal includes,

[0030] determining a confidence difference between the view angle feature corresponding to the confidence view angle in the pork image and the corresponding confidence view angle sample feature;

[0031] if the confidence difference is greater than or equal to a preset difference threshold, determining that the pork image is abnormal;

[0032] if the confidence difference is less than the preset difference threshold, determining that the pork image is not abnormal.

[0033] Further, the process of verifying the pork image without abnormalities by the abnormality identification model includes,

[0034] extracting the pork image without abnormalities at a preset sampling ratio;

[0035] inputting the pork image without abnormalities into a pre-trained abnormality identification model for analysis to verify whether the pork image without abnormalities is still determined to be not abnormal.

[0036] Further, the process of determining whether to update the confidence view angle sample feature based on the verification result includes,

[0037] if the pork image is verified to be abnormal, updating the confidence view angle sample feature.

[0038] Further, the process of updating the confidence view angle sample feature includes,

[0039] reacquiring a plurality of pork image samples corresponding to the label information to obtain a pork image set;

[0040] performing confidence view angle analysis on the pork image set to determine a new confidence view angle sample feature;

[0041] the confidence view angle sample feature is used as the updated confidence view angle sample feature.

[0042] On the other hand, the present application provides a system for applying a pork traceability management method based on big data, comprising:

[0043] a collection module configured to acquire a plurality of pork image samples and corresponding label information, wherein the label information includes the origin, part, and slaughter time interval corresponding to the pork image samples;

[0044] a confidence visual angle construction module connected with the collection module, configured to classify pork image samples based on label information to obtain a plurality of pork image sets, and perform confidence visual angle analysis on each pork image set, including determining a confidence visual angle for the label information based on differences in texture visual angle features and differences in chroma visual angle features between pork image samples, and synchronously constructing confidence visual angle sample features;

[0045] a processing module connected with the confidence visual angle construction module, configured to, in response to collecting production batch information, pork images, and label information, call the confidence visual angle associated with the label information, and process the pork images, including,

[0046] extracting visual angle features corresponding to the confidence visual angle in the pork images and matching the visual angle features with the corresponding confidence visual angle sample features to determine whether the pork images are abnormal; or analyzing the pork images through a pre-trained abnormality recognition model to determine whether the pork images are abnormal;

[0047] a verification module connected with the processing module, configured to perform spot-check verification on non-abnormal pork images through the abnormality recognition model, and determine whether to update the confidence visual angle sample features based on a verification result;

[0048] a traceability module connected with the collection module, the confidence visual angle construction module, and the processing module, respectively, and configured to determine that a production batch corresponding to an abnormal pork image needs to be traced.

[0049] Compared with the prior art, the present application acquires pork image samples and corresponding label information, classifies the samples based on the label information, and constructs confidence visual angle sample features, determines the confidence visual angle by analyzing differences in texture visual angle features and chroma visual angle features, and then processes the collected production batch information, pork images, and label information, including, in the processing stage, extracting visual angle features corresponding to the confidence visual angle in the pork images, matching the visual angle features with the confidence visual angle sample features to determine whether the images are abnormal, and tracing the production batch corresponding to the abnormal images. The present application improves the processing efficiency and accuracy of the traceability system by combining confidence visual angle analysis and an abnormality recognition model.

[0050] Especially, the application performs confidence visual angle analysis on each of the pork image sets. In actual situations, due to pork of different origins, different parts, and different freshness, the appearance of the pork in a normal state is already quite different. For example, the texture of tenderloin and pork belly is significantly different, and the color of pork 1 hour and 24 hours after slaughter is also different. If a unified and complex model is used to detect all types of pork, the model needs to learn extremely complex feature boundaries, which will cause the model itself to become large and slow in calculation, and a large amount of labeled data is needed to cover all possibilities, which is prone to waste of resources and inefficiency. However, within the same category, i.e. pork image samples of certain specific origin and certain specific part, always show highly consistent, stable and repeatable visual features. For example, the texture and color of pork of certain origin and certain part show high consistency, which can be used as a basis for determining the abnormality of pork. The data is highly representative. In some cases, the texture and color of pork of certain origin and certain part already have certain differences, so directly using texture or color to determine the abnormality confidence of pork is low. Based on this, the application sets a specific confidence visual angle for specific label information, which facilitates subsequent selection of analysis methods for pork images based on label information, thereby reducing unnecessary calculations and improving the processing efficiency and accuracy of the traceability system.

[0051] Especially, the application processes pork images by calling the confidence visual angle associated with the label information. Since pork image samples with a confidence visual angle show high consistency and stability in texture and color characteristics, by constructing a confidence visual angle sample feature, determining the mean of the confidence visual angle sample feature, and calculating the difference between the sample image to be detected and the confidence visual angle sample feature, the difference between the sample image and the normal sample can be quantified, and these stable and repeatable visual features can be more accurately captured to determine that the sample image is abnormal. However, not all pork image samples have a confidence visual angle. For those sample images that do not have a confidence visual angle, such as pork of certain special parts or samples that show large variations at a certain slaughter time, the differences in texture and color visual angle features are large, and it is difficult to detect them through a confidence visual angle. Therefore, it is necessary to analyze these sample images in combination with a pre-trained abnormality recognition model to ensure the comprehensiveness and accuracy of the detection. By combining confidence visual angle analysis and an abnormality recognition model, more comprehensive and accurate abnormality detection can be performed, thereby reducing unnecessary calculations and improving the processing efficiency and accuracy of the traceability system.

[0052] Especially, the application verifies the pork image without anomaly through the anomaly identification model, and determines whether to update the confidence visual angle sample feature based on the verification result. In the actual production environment, due to factors such as the source of pork, feed, season, and slaughtering process, slow changes may occur over time, which may lead to deviations or failure of the confidence visual angle sample feature in actual application. For example, as time goes by, the environment changes or the sample feature naturally varies, the original confidence visual angle sample feature may no longer be completely applicable. Through the verification, the difference between the sample image and the normal sample can be quantified, and these changes can be captured in time, so that it can be more accurately determined whether the sample image is abnormal. Therefore, by verifying the sample image without anomaly, the reliability of the confidence visual angle sample feature can be more comprehensively and accurately evaluated, potential problems can be found and corrected in time, thereby reducing unnecessary calculation and improving the processing efficiency and accuracy of the traceability system. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure 1 A step schematic diagram of the pork traceability management method based on big data of the embodiment of the application;

[0054] Figure 2 A logic determination diagram for determining the confidence visual angle of the tag information of the embodiment of the application;

[0055] Figure 3 A logic block diagram for processing the pork image of the embodiment of the application;

[0056] Figure 4 A logic determination diagram for determining whether to update the confidence visual angle sample feature of the embodiment of the application. DETAILED DESCRIPTION

[0057] In order to make the objects and advantages of the application clearer, the application will be further described below with reference to the embodiments. It should be understood that the specific embodiments described herein are only used to explain the application and do not limit the application.

[0058] The preferred embodiments of the application will be described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are only used to explain the technical principles of the application and do not limit the protection scope of the application.

[0059] In addition, it should be further pointed out that in the description of the application, unless otherwise explicitly specified and limited, the term "connection" should be understood broadly, for example, it can be fixed connection, detachable connection, or integral connection; it can be mechanical connection, electrical connection; it can be direct connection, indirect connection through an intermediate medium, or internal communication of two elements. Those skilled in the art can understand the specific meaning of the above-mentioned term in the application according to the specific circumstances.

[0060] Referring to Figure 1 As shown in the step schematic diagram of the pork traceability management method based on big data of the embodiment, the pork traceability management method based on big data of the embodiment comprises:

[0061] Obtain a plurality of pork image samples and corresponding label information, the label information comprising a producing area, a part and a slaughtering time interval corresponding to the pork image samples;

[0062] Classify the pork image samples based on the label information to obtain a plurality of pork image sets, and perform confidence perspective analysis on each pork image set, comprising determining a confidence perspective for the label information based on differences in texture perspective features and differences in chrominance perspective features between the pork image samples, and simultaneously constructing confidence perspective sample features;

[0063] In response to collecting production batch information, pork images and label information, calling the confidence perspective associated with the label information, processing the pork images, comprising,

[0064] Matching the perspective features corresponding to the confidence perspective in the pork images with the corresponding confidence perspective sample features to determine whether the pork images are abnormal; or, analyzing the pork images through a pre-trained abnormality recognition model to determine whether the pork images are abnormal;

[0065] Verifying the abnormality recognition model for the non-abnormal pork images, and determining whether to update the confidence perspective sample features based on the verification result;

[0066] Determining that the production batch corresponding to the abnormal pork images needs to be traced;

[0067] The confidence perspective comprises a texture confidence perspective and a chrominance confidence perspective, and the confidence perspective sample features comprise texture confidence perspective sample features and chrominance confidence perspective sample features.

[0068] In implementation, the manner of obtaining a plurality of pork image samples and corresponding label information is not limited, and can be obtained through an existing database or by deploying a high-definition camera at key production links such as slaughterhouses and meat processing workshops to collect pork image samples in real time. It is only necessary to ensure that the collected image samples have sufficient clarity and representativeness, and can accurately reflect the texture, color and other characteristics of pork, which will not be repeated here.

[0069] The pork image samples should be preprocessed according to a unified standard, including size normalization, background removal, color correction and the like, so as to eliminate the interference of irrelevant variables on subsequent analysis.

[0070] In implementation, the label information includes origin, part and slaughter duration interval corresponding to the pork image sample, wherein the slaughter duration interval is a time period from the beginning of slaughter to the collection of the image sample, and is usually divided in units of hours, for example, the slaughter duration interval can be divided into different category sets such as [0 hours, 5 hours], [6 hours, 11 hours], [11 hours, 16 hours] and the like.

[0071] Specifically, the classification of the pork image sample based on the label information obtains a plurality of pork image sets, and the process of confidence perspective analysis of each pork image set includes,

[0072] pork image samples with the same label information are classified into the same category to obtain a pork image set;

[0073] determining the texture feature and the color feature of the pork image sample in each pork image set;

[0074] extracting the lean meat area of each pork image sample, determining the texture interval variance of the lean meat area between the pork image samples, and determining the texture interval variance as the difference of the texture perspective feature;

[0075] determining the color variance between the pork image samples, and determining the color variance as the difference of the color perspective feature.

[0076] In implementation, for the convenience of calculation, the texture feature and the color feature need to be normalized to the range of [0, 1], and then the variance is calculated to ensure the comparability of the variance.

[0077] In implementation, the way of determining the texture feature and the color feature of the pork image sample in each pork image set is not limited, which can be extracted by image processing technology such as gray level co-occurrence matrix and the like, and then the average interval between a single texture and the most adjacent texture is determined, and then the variance corresponding to the average interval is calculated to obtain the texture interval variance, and the color feature can be extracted by color histogram or other methods, as long as the features can be accurately extracted and quantified.

[0078] In implementation, the method of extracting the lean meat area is not limited, which can adopt an algorithm based on image segmentation, such as threshold segmentation, edge detection or region growing, as long as the lean meat area can be accurately extracted, and details are not repeated.

[0079] Please refer to Figure 2 The logic decision diagram for determining the confidence perspective of the label information of the embodiment of the application is shown in the figure, and specifically, the process of determining the confidence perspective of the label information based on the difference of the texture perspective feature and the difference of the color perspective feature between the pork image samples includes,

[0080] If the difference of the texture view angle feature is less than a preset texture difference threshold, it is determined that the label information has a texture confidence view angle, and texture confidence sample features are constructed synchronously.

[0081] If the difference of the chroma view angle feature is less than a preset chroma difference threshold, it is determined that the label information has a chroma confidence view angle, and chroma confidence sample features are constructed synchronously.

[0082] In implementation, the purpose of setting the texture difference threshold is to screen out the case that the texture features of the pork image samples are highly close. Therefore, the texture difference threshold should not be too high. Since the difference of the texture view angle feature is essentially a variance, in order to reflect the case that the variance is small and the data is stable, the texture difference threshold is in the interval [0.1, 0.15], and in implementation, 0.1 is preferred.

[0083] In implementation, the purpose of setting the chroma difference threshold is to screen out the case that the chroma features of the pork image samples are highly close. Therefore, the chroma difference threshold should not be too high. Since the difference of the chroma view angle feature is essentially a variance, in order to reflect the case that the variance is small and the data is stable, the chroma difference threshold is in the interval [0.05, 0.1], and in implementation, 0.07 is preferred.

[0084] In actual situations, since the pork of different origins, different parts and different freshness has a great difference in appearance under normal conditions, for example, the texture of tenderloin and pork belly has a significant difference, and the color of pork 1 hour and 24 hours after slaughter is also different, if a unified and complex model is used to detect all types of pork, the model needs to learn extremely complex feature boundaries, which will cause the model itself to become large and slow in calculation, and a large amount of labeled data is needed to cover all possibilities, which is easy to cause resource waste and inefficiency. Under the same category, i.e. pork image samples of certain specific origin and specific part always present highly consistent, stable and repeatable visual features. For example, the texture and chroma of pork of certain origin and certain part present high consistency, which can be used as a basis for determining pork abnormality. The data is highly representative. In some cases, the texture and chroma of pork of certain origin and certain part have certain differences. Therefore, directly using texture or chroma to determine pork abnormality confidence is low. Based on this, the present application sets specific confidence view angles for specific label information, which facilitates subsequent selection of analysis methods for pork images based on label information, thereby reducing unnecessary calculation and improving the processing efficiency and accuracy of the traceability system.

[0085] Specifically, the process of synchronously constructing the confidence view angle sample features includes,

[0086] extracting a mean value of fiber texture interval of lean meat area in a pork image sample in each of the pork image set, and determining the mean value of the fiber texture interval as a texture confidence view sample feature;

[0087] extracting a mean value of chroma between pork image samples in each of the pork image set, and determining the mean value of the chroma as a chroma confidence view sample feature.

[0088] In implementation, the extraction of the mean value of the chroma is not limited, and the mean value of each channel can be calculated by converting the image to different color spaces or by other methods, as long as the features can be accurately extracted and quantified.

[0089] In implementation, the confidence view sample feature includes one or both of the texture confidence view sample feature and the chroma confidence view sample feature.

[0090] Please refer to Figure 3 The logic block diagram of the application embodiment for processing the pork image is shown, and specifically, the process of calling the confidence view associated with the label information for processing the pork image includes,

[0091] If the label information has a confidence view, the view features corresponding to the confidence view in the pork image are extracted and matched with the corresponding confidence view sample features to determine whether the pork image has an abnormality.

[0092] If the label information does not have a confidence view, the pork image is analyzed by the pre-trained abnormality recognition model to determine whether the pork image has an abnormality.

[0093] In implementation, the form of the pre-trained abnormality recognition model is not limited, and an existing open-source image processing model for recognizing pork abnormalities can be used, or a corresponding image processing module can be trained as an abnormality recognition model, for example, an image processing model with a neural network architecture is used, and pork images with abnormalities and pork images without abnormalities are collected as training samples to train an image processing model capable of recognizing abnormalities in pork images as an abnormality recognition model, which will not be repeated.

[0094] Specifically, the process of extracting the view features corresponding to the confidence view in the pork image and matching them with the corresponding confidence view sample features to determine whether the pork image has an abnormality includes,

[0095] determining the confidence difference between the view features corresponding to the confidence view in the pork image and the corresponding confidence view sample features; it can be understood that the confidence difference is the difference between the view features corresponding to the confidence view and the corresponding confidence view sample features;

[0096] If the confidence difference value is greater than or equal to a preset difference threshold value, it is determined that the pork image is abnormal.

[0097] If the confidence difference value is less than the preset difference threshold value, it is determined that the pork image is not abnormal.

[0098] It can be understood that the confidence perspective includes a texture confidence perspective and a chroma confidence perspective, and accordingly, a corresponding difference threshold value needs to be matched, which includes a texture difference threshold value and a chroma difference threshold value.

[0099] In implementation, if the confidence difference value corresponding to the texture confidence perspective is greater than or equal to the texture difference threshold value, it is determined that the pork image is abnormal.

[0100] If the confidence difference value corresponding to the chroma confidence perspective is greater than or equal to the chroma difference threshold value, it is determined that the pork image is abnormal.

[0101] In implementation, the purpose of setting the difference threshold value is to reflect the case of abnormality, and the difference threshold value is pre-set, wherein, the abnormal pork image samples and the normal pork image samples corresponding to the same label information are pre-collected, the confidence difference value between the confidence perspective features of the abnormal pork image samples and the confidence perspective features of the normal pork image samples is determined, including the texture confidence difference value and the chroma confidence difference value.

[0102] The mean value of the texture confidence difference value and the mean value of the chroma confidence difference value are solved.

[0103] The texture difference threshold value in the difference threshold value is set as the product of the mean value of the texture confidence difference value and an error coefficient.

[0104] The chroma difference threshold value in the difference threshold value is set as the product of the mean value of the chroma confidence difference value and an error coefficient.

[0105] The error coefficient is selected in the interval [1.15, 1.25], and in implementation, it is preferably 1.15.

[0106] The application processes pork image by calling the confidence visual angle associated with the label information. Since pork image samples with confidence visual angle show high consistency and stability in texture and chroma characteristics, the difference between the sample image and the normal sample can be quantified by determining the mean value of the confidence visual angle sample characteristics and calculating the difference between the sample image to be detected and the confidence visual angle sample characteristics, which can more accurately capture these stable and repeatable visual features and determine that the sample image is abnormal. However, not all pork image samples have confidence visual angle. For those sample images without confidence visual angle, such as pork of certain special parts or sample images showing large variation under certain slaughter time, the difference in texture and chroma visual angle characteristics is large, and it is difficult to detect by confidence visual angle. Therefore, it is necessary to analyze these sample images by combining a pre-trained abnormality recognition model to ensure the comprehensiveness and accuracy of detection. By combining confidence visual angle analysis and abnormality recognition model, more comprehensive and accurate abnormality detection can be performed, thereby reducing unnecessary calculation and improving the processing efficiency and accuracy of the traceability system.

[0107] Referring to Figure 4 As shown in the logical determination diagram for determining whether to update the confidence visual angle sample characteristics of the embodiment, specifically, the process of verifying the no-abnormal pork image by the abnormality recognition model includes,

[0108] extracting the no-abnormal pork image according to the preset sampling ratio;

[0109] inputting the no-abnormal pork image into the pre-trained abnormality recognition model for analysis to verify whether the no-abnormal pork image is still determined to be no-abnormal.

[0110] In implementation, the sampling ratio is selected within the interval [25%, 50%], and in implementation, it is preferably 30%.

[0111] Specifically, the process of determining whether to update the confidence visual angle sample characteristics based on the verification result includes,

[0112] if the verified pork image is abnormal, update the confidence visual angle sample characteristics.

[0113] Specifically, the process of updating the confidence visual angle sample characteristics includes,

[0114] reacquire a plurality of pork image samples corresponding to the label information to obtain a pork image set;

[0115] perform confidence visual angle analysis on the pork image set to determine new confidence visual angle sample characteristics;

[0116] The confidence perspective sample feature is taken as an updated confidence perspective sample feature.

[0117] In implementation, in order to ensure that the updated confidence perspective sample feature can more accurately reflect the feature distribution of the current sample, and in the subsequent can give priority to these updated confidence perspective sample features for re-inspection.

[0118] The application verifies the non-abnormal pork image by the abnormal identification model, and determines whether to update the confidence perspective sample feature based on the verification result. In actual production environment, due to factors such as source of pork, feed, season, slaughtering process and the like, slow changes may occur over time, which may lead to deviation or failure of the confidence perspective sample feature in actual application. For example, with the passage of time, changes in the environment or natural variation of sample features, the original confidence perspective sample feature may no longer be completely applicable. Through the verification of the re-inspection, the difference between the sample image and the normal sample can be quantified, and these changes can be captured in time, so as to more accurately determine whether the sample image is abnormal. Therefore, by verifying the non-abnormal sample image, the reliability of the confidence perspective sample feature can be more comprehensively and accurately evaluated, potential problems can be found and corrected in time, thereby reducing unnecessary calculation and improving the processing efficiency and accuracy of the traceability system.

[0119] It can be understood that for the pork image with abnormality, the corresponding generation batch information can be determined for traceability.

[0120] Specifically, the application embodiment also provides a system of a pork traceability management method based on big data, comprising:

[0121] The acquisition module is used to obtain a plurality of pork image samples and corresponding label information, and the label information includes the origin, part and slaughtering time interval corresponding to the pork image sample;

[0122] The confidence perspective construction module is connected with the acquisition module, and is used to classify the pork image samples based on the label information to obtain a plurality of pork image sets, and perform confidence perspective analysis on each pork image set, including determining the confidence perspective for the label information based on the differences in texture perspective features and the differences in color perspective features between the pork image samples, and synchronously constructing the confidence perspective sample feature;

[0123] The processing module is connected with the confidence perspective construction module, and in response to the acquisition of the production batch information, the pork image and the label information, the confidence perspective associated with the label information is called to process the pork image, including,

[0124] The confidence view angle feature in the pork image is matched with the corresponding confidence view angle sample feature to determine whether the pork image is abnormal; or, the pork image is analyzed by the pre-trained abnormality recognition model to determine whether the pork image is abnormal.

[0125] A verification module is connected with the processing module, and the abnormality recognition model is used to verify the pork image without abnormality, and whether the confidence view angle sample feature is updated is determined based on the verification result;

[0126] A traceability module is connected with the acquisition module, the confidence view angle construction module and the processing module respectively, and is used to determine whether the production batch corresponding to the abnormal pork image needs to be traced.

[0127] In implementation, the configuration of the acquisition module is not limited, which can be constituted by an image acquisition device such as an industrial camera, a high-definition camera, etc., or image samples and label information can be obtained from a remote database through a network interface, as long as the acquired image samples are clear and complete and the label information is accurate, which will not be repeated here.

[0128] In implementation, the structures of the confidence view angle construction module, the processing module, the verification module and the traceability module are not limited, which can be constituted by a logic component or a combination of logic components, and the logic component includes a field programmable processor, a computer or a microprocessor in the computer, which will not be repeated here.

[0129] So far, the technical solutions of the present application have been described in combination with the preferred embodiments shown in the drawings, but those skilled in the art can easily understand that the protection scope of the present application is obviously not limited to these specific embodiments. Those skilled in the art can make equivalent changes or replacements to related technical features without departing from the principles of the present application, and the technical solutions after the changes or replacements will fall within the protection scope of the present application.

Claims

1. A big data-based pork traceability management method, characterized by, The method comprises the following steps: acquire a plurality of pork image samples and corresponding label information, the label information includes the origin, the part and the slaughter time interval corresponding to the pork image sample; based on the label information, the pork image samples are classified to obtain a plurality of pork image sets, and confidence visual angle analysis is performed on each pork image set, including determining the confidence visual angle for the label information based on the difference in texture visual angle features and the difference in color visual angle features between the pork image samples, and simultaneously constructing confidence visual angle sample features; in response to collecting production batch information, pork image and label information, calling the confidence visual angle associated with the label information, processing the pork image, including, if the label information has a confidence visual angle, extract the visual angle features corresponding to the confidence visual angle in the pork image and match the corresponding confidence visual angle sample features to determine whether the pork image is abnormal; or, if the label information has no confidence visual angle, analyze the pork image through a pre-trained abnormality recognition model to determine whether the pork image is abnormal; through the abnormality recognition model, the abnormal pork image is randomly inspected and verified, and based on the verification result, it is determined whether to update the confidence visual angle sample features; determine that the production batch corresponding to the abnormal pork image needs to be traced; wherein the confidence visual angle includes texture confidence visual angle and color confidence visual angle, and the confidence visual angle sample features include texture confidence visual angle sample features and color confidence visual angle sample features.

2. The big data-based pork traceability management method of claim 1, characterized in that, The process of classifying pork image samples based on label information to obtain a plurality of pork image sets and performing confidence visual angle analysis on each pork image set comprises: pork image samples with the same label information are classified into the same category to obtain pork image sets; determine the texture features and color features of the pork image samples in each pork image set; extract the lean meat area of each pork image sample, determine the texture interval variance of the lean meat area between the pork image samples, and determine the texture interval variance as the difference in texture visual angle features; determine the color difference between the pork image samples, and determine the color difference as the difference in color visual angle features.

3. The big data-based pork traceability management method of claim 2, characterized in that, The process of determining the confidence visual angle for the label information based on the difference in texture visual angle features and the difference in color visual angle features between the pork image samples comprises: if the difference in texture visual angle features is less than a preset texture difference threshold, it is determined that the label information has a texture confidence visual angle, and texture confidence sample features are simultaneously constructed; if the difference in color visual angle features is less than a preset color difference threshold, it is determined that the label information has a color confidence visual angle, and color confidence sample features are simultaneously constructed.

4. The big data-based pork traceability management method of claim 1, wherein, The process of simultaneously constructing confidence visual angle sample features comprises: extract the average fiber texture interval of the lean meat area in each pork image sample in each pork image set, and determine the average fiber texture interval as the texture confidence visual angle sample features; extract the color average value between the pork image samples in each pork image set, and determine the color average value as the color confidence visual angle sample features.

5. The big data-based pork traceability management method of claim 4, wherein, The process of matching the view angle feature corresponding to the confidence view angle in the extracted pork image with the corresponding confidence view angle sample feature to determine whether the pork image is abnormal includes, determining the confidence difference between the view angle feature corresponding to the confidence view angle in the pork image and the corresponding confidence view angle sample feature; if the confidence difference is greater than or equal to a preset difference threshold, it is determined that the pork image is abnormal; if the confidence difference is less than the preset difference threshold, it is determined that the pork image is normal.

6. The big data-based pork traceability management method of claim 5, wherein, The process of verifying the normal pork image by the abnormality identification model includes, extracting the normal pork image at a preset sampling rate; inputting the normal pork image into the pre-trained abnormality identification model for analysis to verify whether the normal pork image is still determined to be normal.

7. The big data-based pork traceability management method of claim 6, wherein, The process of determining whether to update the confidence view angle sample feature based on the verification result includes, if the pork image is verified to be abnormal, updating the confidence view angle sample feature. 8.The big data-based pork traceability management method of claim 7, wherein, The process of updating the confidence view angle sample feature includes, reacquiring a plurality of pork image samples corresponding to the label information to obtain a pork image set; performing confidence view angle analysis on the pork image set to determine new confidence view angle sample features; the confidence view angle sample features are used as updated confidence view angle sample features.

9. A system for applying the pork traceability management method based on big data according to any one of claims 1 to 8, characterized in that, includes, a collection module configured to acquire a plurality of pork image samples and corresponding label information, the label information including the origin, part, and slaughter time interval corresponding to the pork image samples; a confidence view angle construction module connected to the collection module and configured to classify the pork image samples based on the label information to obtain a plurality of pork image sets, and perform confidence view angle analysis on each pork image set, including determining the confidence view angle for the label information based on the differences in texture view angle features and color view angle features between the pork image samples, and simultaneously constructing confidence view angle sample features; a processing module connected to the confidence view angle construction module, and configured to respond to the collection of production batch information, pork images, and label information, call the confidence view angle associated with the label information, and process the pork images, including, if the label information has a confidence view angle, matching the view angle feature corresponding to the confidence view angle in the pork image with the corresponding confidence view angle sample feature to determine whether the pork image is abnormal; or, if the label information does not have a confidence view angle, analyzing the pork image by a pre-trained abnormality identification model to determine whether the pork image is abnormal; a verification module connected to the processing module, configured to verify the normal pork image by the abnormality identification model, and determine whether to update the confidence view angle sample feature based on the verification result; a traceability module connected to the collection module, confidence view angle construction module, and processing module, respectively, and configured to determine that the production batch corresponding to the abnormal pork image needs to be traced.

Images