A method and system for the production of selenium-enriched animal products

Abstract

The present application relates to the technical field of livestock product production, and relates to a preparation method for realizing selenium-rich livestock products, which comprises the following steps: extracting a standard breeding video set and a corresponding standard breeding text set from selenium-rich breeding data; extracting a livestock state feature set from the standard breeding video set; extracting a feed formula feature set, a feeding plan feature set and a selenium-rich conversion feature set from the standard breeding text set; training a preset selenium-rich feeding model into a selenium-rich conversion model by using the livestock state feature set, the feed formula feature set, the feeding plan feature set and the selenium-rich conversion feature set; acquiring a real-time livestock video, extracting real-time livestock state features from the real-time livestock video, analyzing a real-time feeding scheme of the real-time livestock state features by using the selenium-rich conversion model, and preparing the selenium-rich livestock products according to the real-time feeding scheme. The present application also provides a preparation system for realizing the selenium-rich livestock products. The present application can improve the biological conversion rate of the preparation of the selenium-rich livestock products.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of livestock product production, and particularly relates to a preparation method and system for realizing selenium-rich livestock products. BACKGROUND

[0002] Selenium is an important trace element and plays an important role in human health, including participating in antioxidant reactions, immune regulation, promoting thyroid function, etc. Selenium-rich livestock products are one of the main sources of selenium elements for the human body. Selenium-rich livestock products refer to products with high selenium content in livestock products after adding components rich in selenium elements in animal feed and being digested and absorbed by animals.

[0003] Existing preparation methods for selenium-rich livestock products are mostly based on artificial formula, that is, the formula of the feed is adjusted according to the suggestions of veterinarians or animal husbandry experts, and the selenium-rich livestock products are prepared by feeding livestock with the feed with the adjusted formula. In actual application, the preparation method based on artificial formula needs real-time determination and judgment by artificial, and the type of the formula is low, the flexibility is poor, and the biological conversion rate of selenium elements is not considered, which may result in a low biological conversion rate when preparing selenium-rich livestock products. SUMMARY

[0004] The present application provides a preparation method and system for realizing selenium-rich livestock products, which mainly aims to solve the problem of low biological conversion rate when preparing selenium-rich livestock products.

[0005] To achieve the above-mentioned purpose, the present application provides a preparation method for realizing selenium-rich livestock products, which comprises:

[0006] The selenium-rich breeding data is split into a period breeding data set according to a breeding period, the period breeding data set is cleaned into a standard breeding data set, and a standard breeding video set and a corresponding standard breeding text set are extracted from the standard breeding data set;

[0007] Key frames are extracted from the standard breeding video set to obtain a breeding livestock picture group set, livestock type feature sets, livestock size feature sets, and livestock stage feature sets are respectively extracted from the breeding livestock picture group set, and the livestock type feature sets, the livestock size feature sets, and the livestock stage feature sets are fused into a livestock state feature set, wherein the key frames are extracted from the standard breeding video set to obtain the breeding livestock picture group set, which comprises: selecting a standard breeding video in the standard breeding video set as a target standard breeding video one by one, and cutting the target standard breeding video into a standard breeding picture set by using a preset video segmentation tool; a target comparison breeding picture group is composed of two continuous standard breeding pictures in the standard breeding picture set one by one, and a variation rate algorithm is used to calculate the variation rate corresponding to the target comparison breeding picture group as follows:

[0008]

[0009] Wherein, S refers to the variation rate, m refers to the pixel length of each standard breeding picture in the target contrast breeding picture set, n refers to the pixel width of each standard breeding picture in the target contrast breeding picture set, a is a preset first variation weight coefficient, b is a preset second variation weight coefficient, A max refers to the maximum gray value of the pixel in the first standard breeding picture in the target contrast breeding picture set, A min refers to the minimum gray value of the pixel in the first standard breeding picture in the target contrast breeding picture set, B max refers to the maximum gray value of the pixel in the second standard breeding picture in the target contrast breeding picture set, B min refers to the minimum gray value of the pixel in the second standard breeding picture in the target contrast breeding picture set, i refers to the i-th, j refers to the j-th, A i,j refers to the gray value of the pixel with the pixel coordinate (i, j) in the first standard breeding picture in the target contrast breeding picture set, B i,j refers to the gray value of the pixel with the pixel coordinate (i, j) in the second standard breeding picture in the target contrast breeding picture set; a livestock confidence set of the variation breeding picture set is calculated by using a target detection method according to the variation rate; a breeding livestock picture set is screened from the variation breeding picture set according to the livestock confidence set, and all the breeding livestock picture sets are collected into a breeding livestock picture set set;

[0010] Text feature extraction, feature conversion and feature segmentation operations are sequentially performed on the standard breeding text set to obtain a feed formula feature set, a feeding plan feature set and a selenium enrichment conversion feature set.

[0011] The livestock state feature set, the feed formula feature set, the feeding plan feature set and the selenium enrichment conversion feature set are used to train a preset selenium enrichment feeding model into a selenium enrichment conversion model.

[0012] Real-time livestock video is acquired, real-time livestock state features are extracted from the real-time livestock video, the real-time feeding scheme corresponding to the real-time livestock state features is analyzed by using the selenium enrichment conversion model, and the preparation of the selenium livestock product is performed according to the real-time feeding scheme.

[0013] Optionally, the cleaning of the periodic breeding data set into a standard breeding data set comprises:

[0014] The repeated breeding data and the default breeding data are sequentially screened out from the periodic breeding data set to obtain an initial breeding data set.

[0015] splitting the initial farming data set into a text farming data set and a video farming data set according to a data format;

[0016] performing unit normalization and outlier rejection operation on each text farming data in the text farming data set in sequence to obtain a standard text farming data set;

[0017] performing video transcoding on each video farming data in the video farming data set to obtain a standard video farming data set;

[0018] pools the standard text farming data set and the standard video farming data set into a standard farming data set.

[0019] Optionally, the method for calculating the livestock confidence set of the variable farming picture set by using target detection comprises:

[0020] selects a variable farming picture in the variable farming picture set as a target variable farming picture, extracts a variable farming contour from the target variable farming picture, and generates a contour bounding box according to the variable farming contour;

[0021] performs filtering operation on the target variable farming picture by using a texture feature extraction algorithm to obtain a target farming texture picture;

[0022]

[0023] wherein G(x, y) is the gray value of a pixel point with coordinates (x, y) in the target farming texture picture, exp() is an exponential function, is the horizontal coordinate of a pixel point in the target variable farming picture, is the vertical coordinate of a pixel point in the target variable farming picture, σ is the standard deviation of a Gaussian function, o is an imaginary symbol, π is a circular constant, f is a frequency parameter, x is the horizontal coordinate of a pixel point in the target farming texture picture, and φ is a phase parameter;

[0024] performs multi-level convolution and multi-level pooling operation on the target farming texture picture to obtain a target farming feature;

[0025] calculates a predicted bounding box and a livestock frame probability of the target farming feature by using a pre-trained livestock target detection model;

[0026] calculates a livestock confidence by using a confidence algorithm according to the predicted bounding box, the contour bounding box, and the livestock frame probability, and pools all the livestock confidences into a livestock confidence set:

[0027]

[0028] Wherein, O refers to the livestock confidence, t refers to the livestock frame probability, Y is the predicted bounding box, L is the contour bounding box, ∩ is the intersection symbol, and ∪ is the union symbol.

[0029] Optionally, the extracting the livestock species feature set, the livestock size feature set, and the livestock stage feature set from the livestock image group set respectively comprises:

[0030] Selecting the livestock image group in the livestock image group set as a target livestock image group one by one, and extracting a livestock contour group from the target livestock image group;

[0031] Splitting the target livestock image group into a livestock image group and a background image group according to the livestock contour group;

[0032] Performing reference object identification on the background image group to obtain a reference object image group, and calculating a livestock size feature according to the reference object image group and the livestock image group;

[0033] Extracting a livestock texture feature, a livestock color feature, and a livestock joint feature from the livestock image group in turn, and collecting the livestock texture feature, the livestock color feature, and the livestock joint feature into a livestock species feature;

[0034] Extracting a livestock abdomen feature and a livestock posture feature from the livestock image group in turn, and collecting the livestock abdomen feature and the livestock posture feature into a livestock stage feature;

[0035] Collecting all the livestock size features into a livestock size feature set, collecting all the livestock species features into a livestock species feature set, and collecting all the livestock stage features into a livestock stage feature set.

[0036] Optionally, the fusing the livestock species feature set, the livestock size feature set, and the livestock stage feature set into a livestock state feature set comprises:

[0037] Selecting the livestock species feature in the livestock species feature set as a target livestock species feature one by one, and screening a livestock size feature corresponding to the target livestock species feature from the livestock size feature set as a target livestock size feature;

[0038] Selecting a livestock stage feature corresponding to the target livestock species feature from the livestock stage feature set as a target livestock stage feature;

[0039] Global pooling the target livestock species feature into a dimension-reduced livestock species feature, global pooling the target livestock size feature into a dimension-reduced livestock size feature, and global pooling the target livestock stage feature into a dimension-reduced livestock stage feature;

[0040] Fusing the dimension-reduced livestock species features, the dimension-reduced livestock size features, and the dimension-reduced livestock stage features as dimension features into livestock state features, and collecting all the livestock state features into a livestock state feature set.

[0041] Optionally, the text feature extraction, feature conversion, and feature segmentation operations are sequentially performed on the standard breeding text set to obtain a feed formula feature set, a feeding plan feature set, and a selenium conversion feature set, including:

[0042] Selecting a standard breeding text in the standard breeding text set as a target breeding text one by one, and sequentially performing text segmentation, stop word filtering, and text vectorization operations on the target breeding text to obtain a target breeding text feature;

[0043] Sequentially performing position encoding and attention encoding operations on the target breeding text feature to obtain a target breeding text encoding;

[0044] Sequentially performing residual connection and linear activation operations on the target breeding text encoding to obtain a target text type encoding;

[0045] Decoding a target text type label from the target text type encoding, and splitting the target breeding text into a feed formula text, a feeding plan text, and a livestock product selenium content text according to the target text type label;

[0046] Extracting a feed formula feature from the feed formula text, extracting a feeding plan feature from the feeding plan text, and calculating a selenium conversion feature according to the feed formula text, the feeding plan text, and the livestock product selenium content text;

[0047] Collecting all the feed formula features into a feed formula feature set, collecting all the feeding plan features into a feeding plan feature set, and collecting all the selenium conversion features into a selenium conversion feature set.

[0048] Optionally, the calculating of the selenium conversion feature according to the feed formula text, the feeding plan text, and the livestock product selenium content text includes:

[0049] Extracting a feed ingredient and an ingredient proportion from the feed formula text, and extracting a feed selenium proportion from the feed ingredient and the ingredient proportion;

[0050] Extracting a feeding frequency and a single feeding amount from the feeding plan text, and calculating a total feeding amount according to the feeding frequency and the single feeding amount;

[0051] Multiplying the feed selenium proportion by the total feeding amount to obtain a feed selenium content;

[0052] divide the selenium content of the livestock product by the selenium content of the feed to obtain a selenium enrichment conversion rate, and vectorize the selenium enrichment conversion rate into a selenium enrichment conversion feature.

[0053] Optionally, the training of the preset selenium enrichment feeding model into a selenium enrichment conversion model using the livestock state feature set, the feed formula feature set, the feeding plan feature set, and the selenium enrichment conversion feature set comprises:

[0054] fusing the livestock state feature set, the feed formula feature set, and the feeding plan feature set into a livestock product preparation feature group set;

[0055] calculating a predicted conversion feature set corresponding to the livestock product preparation feature group set using the preset selenium enrichment feeding model;

[0056] calculating a loss value between the predicted conversion feature set and the selenium enrichment conversion feature set;

[0057] determining whether the loss value is greater than a preset loss threshold;

[0058] if yes, updating model parameters of the selenium enrichment feeding model according to the loss value, and returning to the step of calculating the predicted conversion feature set corresponding to the livestock product preparation feature group set using the preset selenium enrichment feeding model;

[0059] if no, taking the updated selenium enrichment feeding model as a selenium enrichment conversion model.

[0060] Optionally, the analysis of the real-time feeding scheme corresponding to the real-time livestock state feature using the selenium enrichment conversion model comprises:

[0061] generating a real-time preparation feature group set corresponding to the real-time state feature using a preset random number algorithm;

[0062] calculating a real-time conversion feature set corresponding to the real-time preparation feature group set using the selenium enrichment conversion model;

[0063] extracting a standard conversion feature from the real-time conversion feature set using a simulated annealing algorithm;

[0064] selecting a real-time preparation feature group corresponding to the standard conversion feature from the real-time preparation feature group set as a target preparation feature group;

[0065] extracting a target feed formula feature and a target feeding plan feature from the target preparation feature group;

[0066] generating a real-time feeding scheme according to the target feed formula feature and the target feeding plan feature.

[0067] To solve the above problems, the application further provides a preparation system for realizing selenium-rich livestock products, which comprises:

[0068] a data splitting module, configured to split selenium-rich breeding data into periodical breeding data sets according to breeding periods, clean the periodical breeding data sets into standard breeding data sets, extract standard breeding video sets and corresponding standard breeding text sets from the standard breeding data sets;

[0069] a video feature extraction module, configured to extract key frames from the standard breeding video sets to obtain a breeding livestock picture group set, extract a livestock type feature set, a livestock size feature set and a livestock stage feature set from the breeding livestock picture group set respectively, and fuse the livestock type feature set, the livestock size feature set and the livestock stage feature set into a livestock state feature set, wherein the video feature extraction module comprises the following steps: selecting a standard breeding video in the standard breeding video set as a target standard breeding video one by one, and cutting the target standard breeding video into a standard breeding picture set by using a preset video segmentation tool; selecting two continuous standard breeding pictures in the standard breeding picture set to form a target comparison breeding picture group one by one, and calculating a variation rate corresponding to the target comparison breeding picture group by using a variation rate algorithm as follows:

[0070]

[0071] wherein S represents the variation rate, m represents the pixel length of each standard breeding picture in the target comparison breeding picture group, n represents the pixel width of each standard breeding picture in the target comparison breeding picture group, a is a preset first variation weight coefficient, and β is a preset second variation weight coefficient, A max represents the maximum gray value of pixels in the first standard breeding picture in the target comparison breeding picture group, A min represents the minimum gray value of pixels in the first standard breeding picture in the target comparison breeding picture group, B max represents the maximum gray value of pixels in the second standard breeding picture in the target comparison breeding picture group, B min represents the minimum gray value of pixels in the second standard breeding picture in the target comparison breeding picture group, i represents the i-th, and j represents the j-th, A i,j represents the gray value of a pixel with a pixel coordinate of (i, j) in the first standard breeding picture in the target comparison breeding picture group, and B i,jis referred to as the gray value of a pixel with pixel coordinates (i, j) in the second standard breeding picture in the target contrast breeding picture set; the variable breeding picture set is screened out from the standard breeding picture set according to the variation rate, and the livestock confidence set of the variable breeding picture set is calculated by using a target detection method; the breeding livestock picture set is screened out from the variable breeding picture set according to the livestock confidence set, and all the breeding livestock picture sets are collected into a breeding livestock picture set set;

[0072] The text feature extraction module is used for sequentially performing text feature extraction, feature conversion and feature segmentation operations on the standard breeding text set, so as to obtain a feed formula feature set, a feeding plan feature set and a selenium conversion feature set.

[0073] The model training module is used for training a preset selenium feeding model into a selenium conversion model by using the livestock state feature set, the feed formula feature set, the feeding plan feature set and the selenium conversion feature set.

[0074] The product preparation module is used for acquiring a real-time livestock video, extracting real-time livestock state features from the real-time livestock video, analyzing a real-time feeding scheme corresponding to the real-time livestock state features by using the selenium conversion model, and preparing a selenium livestock product according to the real-time feeding scheme.

[0075] The livestock state feature set, the feed formula feature set, the feeding plan feature set and the selenium conversion feature set are fused into a livestock state feature set, the state features of the livestock corresponding to each breeding cycle can be extracted from the breeding video, and then different feeding plans can be implemented for livestock in different states, so that the biological conversion rate of the selenium livestock product is improved.

[0076] By utilizing the livestock state feature set, the feed formula feature set, the feeding plan feature set and the selenium-rich conversion feature set, a preset selenium-rich feeding model is trained into a selenium-rich conversion model, so that a relationship model between the influence factors such as the livestock state, the feed formula and the feeding plan in each breeding cycle and the selenium element biological conversion rate of the prepared selenium-rich livestock product is established, and then subsequent targeted improvement of the breeding process and improvement of the biological conversion rate are facilitated, the real-time livestock video is acquired, the real-time livestock state features are extracted from the real-time livestock video, the real-time feeding scheme corresponding to the real-time livestock state features is analyzed by utilizing the selenium-rich conversion model, and the preparation of the selenium-rich livestock product is performed according to the real-time feeding scheme, so that the feeding scheme with the maximum selenium element biological conversion rate is determined according to different states and different physiological characteristics of the livestock, the flexibility of the preparation of the selenium-rich livestock product and the selenium element biological conversion rate are improved. Therefore, the preparation method and system for realizing the preparation of the selenium-rich livestock product can solve the problem of low biological conversion rate during the preparation of the selenium-rich livestock product. BRIEF DESCRIPTION OF DRAWINGS

[0077] Figure 1 A flowchart of the preparation method for realizing the preparation of the selenium-rich livestock product provided by an embodiment of the present application is shown in the figure.

[0078] Figure 2 A flowchart of the preparation method for realizing the preparation of the selenium-rich livestock product provided by an embodiment of the present application is shown in the figure.

[0079] Figure 3 A flowchart of the preparation method for realizing the preparation of the selenium-rich livestock product provided by an embodiment of the present application is shown in the figure.

[0080] Figure 4 A functional module diagram of the preparation system for realizing the preparation of the selenium-rich livestock product provided by an embodiment of the present application is shown in the figure.

[0081] The implementation, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0082] It should be understood that the specific embodiments described herein are merely intended to explain the present application, and are not intended to limit the present application.

[0083] The embodiment of the present application provides a preparation method for realizing selenium-rich livestock products. The execution subject of the preparation method for realizing selenium-rich livestock products includes but is not limited to at least one of electronic devices such as a server, a terminal and the like which can be configured to execute the method provided by the embodiment of the present application. In other words, the preparation method for realizing selenium-rich livestock products can be executed by software or hardware installed in a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to a single server, a server cluster, a cloud server or a cloud server cluster and the like. The server can be a stand-alone server, or a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content distribution networks (CDN), and basic cloud computing services such as big data and artificial intelligence platforms.

[0084] Referring to Figure 1 FIG. 1 is a flowchart of a preparation method for realizing selenium-rich livestock products provided by an embodiment of the present application. In the embodiment, the preparation method for realizing selenium-rich livestock products includes the following steps.

[0085] S1, splitting selenium-rich breeding data into period breeding data sets according to a breeding period, cleaning the period breeding data sets into standard breeding data sets, extracting a standard breeding video set and a corresponding standard breeding text set from the standard breeding data sets.

[0086] In the embodiment of the present application, the breeding period refers to a breeding period for producing selenium-rich livestock products once, the selenium-rich livestock products refer to livestock products rich in selenium elements, such as selenium-rich milk, selenium-rich eggs, selenium-rich beef and the like, and the selenium-rich breeding data refers to record data of preparing selenium-rich livestock products for livestock in a past period of time, including monitoring videos of livestock behavior and various breeding parameters of selenium-rich breeding.

[0087] In detail, the period breeding data set is a data set composed of multiple period breeding data, and each period breeding data is all data recorded in a breeding period for producing selenium-rich livestock products once.

[0088] In the embodiment of the present application, referring to Figure 2 FIG. 2, the cleaning of the period breeding data sets into standard breeding data sets includes the following steps.

[0089] S21, sequentially screening out repeated breeding data and default breeding data from the period breeding data sets to obtain an initial breeding data set;

[0090] S22, splitting the initial breeding data set into a text breeding data set and a video breeding data set according to a data format;

[0091] S23, sequentially performing unit normalization and outlier elimination on each text breeding data in the text breeding data set to obtain a standard text breeding data set;

[0092] S24, performing video transcoding on each video breeding data in the video breeding data set to obtain a standard video breeding data set;

[0093] S25, integrating the standard text breeding data set and the standard video breeding data set into a standard breeding data set.

[0094] Specifically, the repeated breeding data refers to the periodic breeding data in which each data in the periodic breeding data set is the same, and the default breeding data refers to the periodic breeding data in which a data item is missing in the periodic breeding data set. The repeated breeding data can be screened out by using a hash coding or fingerprint algorithm, and the default breeding data can be screened out according to the data cardinality of the periodic breeding data set.

[0095] In detail, the text breeding data set refers to a data set composed of breeding data of which the data type is general text or number, such as recorded feeding ingredients, ingredient proportion in feed, feeding interval period and frequency, and feed volume at each feeding time. The video breeding data set refers to a data set composed of breeding data of which the data type is video, such as a monitoring video for monitoring the growth state of livestock corresponding to a selenium-rich livestock product in one breeding period.

[0096] Specifically, the unit normalization refers to converting the units of salt value data of the same type into the same unit, and the outlier elimination refers to detecting outlier data from the text breeding data and screening out the detected outliers. The outlier elimination can be performed by using a wavelet transform or an outlier detection algorithm, and data interpolation can be performed by using a linear interpolation or a spline interpolation algorithm. The video transcoding on each video breeding data in the video breeding data set to obtain a standard video breeding data set refers to converting the video formats of all video breeding data into a unified format, such as converting AVI and MOV into MP4 format.

[0097] In the embodiment of the application, the standard breeding video set is a collection composed of a plurality of standard breeding videos, and each standard breeding video corresponds to a breeding monitoring video in one periodic breeding data in the periodic breeding data set. The standard breeding text set is a collection composed of a plurality of standard breeding texts, and each standard breeding text corresponds to breeding-related text data recorded in the periodic breeding data set.

[0098] ​In detail, the extracting the standard breeding video set and the corresponding standard breeding text set from the standard breeding data set refers to splitting the breeding data in the breeding data set into standard breeding texts and standard breeding videos according to data types one by one, and collecting all the standard breeding texts into a standard breeding text set and all the standard breeding videos into a standard breeding video set.

[0099] In the embodiment of the present application, by splitting the selenium-rich breeding data into a period breeding data set according to a breeding period, cleaning the period breeding data set into a standard breeding data set, and extracting a standard breeding video set and a corresponding standard breeding text set from the standard breeding data set, the accuracy of the data set used for model training can be improved, and the calculation accuracy of the subsequent model can be improved.

[0100] S2, key frame extraction is performed on the standard breeding video set to obtain a breeding livestock image group set, livestock type feature sets, livestock size feature sets and livestock stage feature sets are extracted from the breeding livestock image group set respectively, and the livestock type feature sets, the livestock size feature sets and the livestock stage feature sets are fused into a livestock state feature set.

[0101] In the embodiment of the present application, the breeding livestock image group set is a set composed of a plurality of breeding livestock image groups, and each breeding livestock image group is a key video frame picture of a standard breeding video in the standard breeding video set.

[0102] In the embodiment of the present application, the key frame extraction on the standard breeding video set to obtain the breeding livestock image group set comprises:

[0103] The standard breeding video in the standard breeding video set is selected as a target standard breeding video one by one, and the target standard breeding video is cut into a standard breeding image set by using a preset video segmentation tool;

[0104] Two continuous standard breeding pictures in the standard breeding image set are selected as a target comparison breeding image group one by one, and a change rate algorithm is used to calculate the change rate corresponding to the target comparison breeding image group as follows:

[0105]

[0106] Wherein, S refers to the change rate, m refers to the pixel length of each standard breeding picture in the target comparison breeding image group, n refers to the pixel width of each standard breeding picture in the target comparison breeding image group, alpha is a preset first change weight coefficient, beta is a preset second change weight coefficient, A max refers to the maximum gray value of the pixels in the first standard breeding picture in the target comparison breeding image group, A minis referred to as the minimum gray value of the pixel in the first standard breeding picture in the target contrast breeding picture set, B max is referred to as the maximum gray value of the pixel in the second standard breeding picture in the target contrast breeding picture set, B min is referred to as the minimum gray value of the pixel in the second standard breeding picture in the target contrast breeding picture set, i is referred to as the i-th, and j is referred to as the j-th, A i,j is referred to as the gray value of the pixel with the pixel coordinates (i, j) in the first standard breeding picture in the target contrast breeding picture set, B i,j is referred to as the gray value of the pixel with the pixel coordinates (i, j) in the second standard breeding picture in the target contrast breeding picture set;

[0107] According to the variation rate, a variation breeding picture set is screened out from the standard breeding picture set, and a livestock confidence set of the variation breeding picture set is calculated by using a target detection method.

[0108] According to the livestock confidence set, a breeding livestock picture set is screened out from the variation breeding picture set, and all the breeding livestock picture sets are collected into a breeding livestock picture set.

[0109] Specifically, the video segmentation tool can be a cv2.VideoCapture and read function of OpenCV or an ffmpeg command of FFmpeg. By calculating the variation rate of the target contrast breeding picture set by using the variation rate algorithm, the difference between pictures can be judged according to the gray value difference of each corresponding coordinate pixel point in the picture and the gray value difference of the pixel extreme value, the picture in which the picture variation occurs is determined, and then the number of processed pictures is reduced, and the algorithm efficiency is improved.

[0110] In detail, the variation breeding picture set with the variation rate greater than a preset variation rate threshold is collected into the variation breeding picture set according to the variation rate from the standard breeding picture set.

[0111] In the embodiment of the application, the livestock confidence set of the variation breeding picture set is calculated by using the target detection method, including:

[0112] The variation breeding pictures in the variation breeding picture set are selected one by one as target variation breeding pictures, variation breeding contours are extracted from the target variation breeding pictures, and a contour boundary box is generated according to the variation breeding contours.

[0113] The target variation breeding picture is subjected to a filtering operation by using a texture feature extraction algorithm, and a target breeding texture picture is obtained.

[0114]

[0115] Wherein, G(x, y) refers to the gray value of the pixel point with coordinates (x, y) in the target breeding texture picture, exp() is the exponential function, is the horizontal coordinate of the pixel point in the target variable breeding picture, is the vertical coordinate of the pixel point in the target variable breeding picture, sigma is the standard deviation of the Gaussian function, o is the imaginary symbol, pi is the circular constant, f is the frequency parameter, x is the horizontal coordinate of the pixel point in the target breeding texture picture, and phi is the phase parameter.

[0116] The target breeding texture picture is subjected to multi-level convolution and multi-level pooling operation to obtain a target breeding feature.

[0117] The pre-trained livestock target detection model is used to calculate the prediction bounding box and livestock frame probability of the target breeding feature.

[0118] The confidence algorithm is used to calculate the livestock confidence according to the prediction bounding box, the contour bounding box and the livestock frame probability, and all the livestock confidences are collected into a livestock confidence set:

[0119]

[0120] Wherein, O refers to the livestock confidence, t refers to the livestock frame probability, Y refers to the prediction bounding box, L refers to the contour bounding box, and is the intersection symbol, is the union symbol.

[0121] In detail, the sobel operator or the canny operator can be used to extract the variable breeding contour from the target variable breeding picture, the contour bounding box is generated according to the variable breeding contour, which refers to constructing a rectangular frame as a contour bounding box outside the variable breeding contour, the target breeding texture picture is obtained by filtering the target variable breeding picture using the texture feature extraction algorithm, the edge, texture and structure features of the image can be extracted in different directions, and the feature details are improved.

[0122] Specifically, the livestock target detection model can be a YOLOv3 model or an EfficientDet model trained by a large number of labeled livestock pictures, the livestock confidence is calculated according to the prediction bounding box, the contour bounding box and the livestock frame probability by using the confidence algorithm, the confidence calculation can be realized by combining the logistic regression function and the degree of overlap between the real bounding box and the prediction bounding box, the accuracy of the confidence is improved, and the breeding livestock picture group is selected from the variable breeding picture set according to the livestock confidence set, which refers to collecting the variable breeding pictures with livestock confidence greater than the preset confidence threshold in the variable breeding picture set into the breeding livestock picture group.

[0123] In detail, the livestock breed feature set is a set composed of multiple livestock breed features, and each livestock breed feature represents the breed and gender of the corresponding livestock of the selenium-rich livestock product in a breeding cycle; the livestock size feature set is a set composed of multiple livestock size features, and each livestock size feature represents the size, whether adult or not, and the like of the corresponding livestock of the selenium-rich livestock product in a breeding cycle; and the livestock stage feature set is a set composed of multiple livestock stage features, and each livestock stage feature represents the growth stage, such as the estrus period and whether pregnant or not, of the corresponding livestock of the selenium-rich livestock product in a breeding cycle.

[0124] In the embodiment of the present application, the livestock breed feature set, the livestock size feature set and the livestock stage feature set are extracted from the breeding livestock image group set respectively, including:

[0125] The breeding livestock image group in the breeding livestock image group set is selected as a target breeding livestock image group one by one, and a livestock contour group is extracted from the target breeding livestock image group.

[0126] The target breeding livestock image group is split into a livestock image group and a background image group according to the livestock contour group.

[0127] A reference object image group is obtained by performing reference object identification on the background image group, and a livestock size feature is calculated according to the reference object image group and the livestock image group.

[0128] A livestock texture feature, a livestock color feature and a livestock joint feature are extracted from the livestock image group in sequence, and the livestock texture feature, the livestock color feature and the livestock joint feature are collected into a livestock breed feature.

[0129] A livestock abdomen feature and a livestock posture feature are extracted from the livestock image group in sequence, and the livestock abdomen feature and the livestock posture feature are collected into a livestock stage feature.

[0130] All the livestock size features are collected into a livestock size feature set, all the livestock breed features are collected into a livestock breed feature set, and all the livestock stage features are collected into a livestock stage feature set.

[0131] Specifically, the method for extracting the livestock contour group from the target breeding livestock image group is the same as the method for extracting the variable breeding contour from the target variable breeding image in step S2, which will not be described here.

[0132] In detail, the YOLOv3 model or the EfficientDet model can be used for reference object identification on the background image set to obtain a reference object image set, the reference object can be a feeding trough, light or a water dispenser, etc., and the livestock size feature calculated according to the reference object image set and the livestock image set refers to that a picture scale set is calculated according to the reference object image set, a livestock size set is calculated according to the picture scale set and the livestock image set, and the median of the livestock size set is taken as the livestock size feature.

[0133] Specifically, the Gabor filtering algorithm or the multi-layer convolution layer can be used to extract the livestock texture feature from the livestock image set, the color histogram method or the color moment algorithm can be used to extract the livestock color feature, the skeleton erosion algorithm or the key point detection algorithm can be used to extract the livestock joint feature, the target detection algorithm can be used to extract the livestock abdomen feature from the livestock image set, and the skeletonization algorithm or the posture estimation network such as OpenPose and AlphaPose can be used to extract the livestock posture feature from the livestock image set.

[0134] Specifically, the livestock species feature set, the livestock size feature set and the livestock stage feature set are fused into a livestock state feature set, which includes:

[0135] The livestock species feature in the livestock species feature set is selected as a target livestock species feature, and the livestock size feature corresponding to the target livestock species feature is selected from the livestock size feature set as a target livestock size feature.

[0136] The livestock stage feature corresponding to the target livestock species feature in the livestock stage feature set is taken as a target livestock stage feature.

[0137] The target livestock species feature is globally pooled into a reduced-dimension livestock species feature, the target livestock size feature is globally pooled into a reduced-dimension livestock size feature, and the target livestock stage feature is globally pooled into a reduced-dimension livestock stage feature.

[0138] The reduced-dimension livestock species feature, the reduced-dimension livestock size feature and the reduced-dimension livestock stage feature are fused into a dimension feature to form a livestock state feature, and all livestock state features are collected into a livestock state feature set.

[0139] In the embodiment of the present application, by extracting key frames from the standard breeding video set, a livestock breeding picture set is obtained, livestock species feature sets, livestock size feature sets and livestock stage feature sets are extracted from the livestock breeding picture set respectively, and the livestock species feature sets, the livestock size feature sets and the livestock stage feature sets are fused into livestock state feature sets, so that the state features of the livestock corresponding to each breeding cycle can be extracted from the breeding video, and different feeding plans can be implemented for livestock in different states, thereby improving the biological conversion rate of selenium-rich livestock products.

[0140] S3, sequentially performing text feature extraction, feature conversion and feature segmentation operations on the standard breeding text set to obtain a feed formula feature set, a feeding plan feature set and a selenium-rich conversion feature set.

[0141] In the embodiment of the present application, the feed formula feature set is a set composed of a plurality of feed formula features, and each feed formula feature corresponds to a formula feature of the feed of the selenium-rich livestock product in a breeding cycle, and the feed formula feature is obtained by encoding the feed composition, the feed component proportion and the like information according to a fixed encoding mode; the feeding plan feature set is a set composed of a plurality of feeding plan features, and each feeding plan feature corresponds to a feeding plan feature of the selenium-rich livestock product in a breeding cycle, and the feeding plan is obtained by encoding the feed feeding frequency, the feed capacity of each feeding and the like information according to a fixed encoding mode.

[0142] In detail, the selenium-rich conversion feature set is a set composed of a plurality of selenium-rich conversion features, and each selenium-rich conversion feature corresponds to the biological conversion rate of the actual selenium element of the selenium-rich livestock product in a breeding cycle.

[0143] In the embodiment of the present application, the sequentially performing text feature extraction, feature conversion and feature segmentation operations on the standard breeding text set to obtain a feed formula feature set, a feeding plan feature set and a selenium-rich conversion feature set comprises:

[0144] The standard breeding text in the standard breeding text set is selected as a target breeding text one by one, and text segmentation, stop word filtering and text vectorization operations are sequentially performed on the target breeding text to obtain a target breeding text feature;

[0145] The target breeding text feature is sequentially subjected to position encoding and attention encoding operations to obtain a target breeding text encoding;

[0146] The target breeding text encoding is sequentially subjected to residual connection and linear activation operations to obtain a target text type encoding;

[0147] decode a target text type label from the target text type encoding, and split the target breeding text into feed formula text, feeding plan text and livestock product selenium content text according to the target text type label;

[0148] extract feed formula features from the feed formula text, extract feeding plan features from the feeding plan text, and calculate a selenium enrichment conversion feature according to the feed formula text, the feeding plan text and the livestock product selenium content text;

[0149] all the feed formula features are collected into a feed formula feature set, all the feeding plan features are collected into a feeding plan feature set, and all the selenium enrichment conversion features are collected into a selenium enrichment conversion feature set.

[0150] In the embodiment of the present application, the bidirectional maximum matching algorithm (Bidirectional Maximum Matching, BMM for short) or the hidden Markov model (Hidden Markov Model, HMM for short) can be used for text segmentation of the target breeding text, the stop word table can be used for stop word filtering of the target breeding text, and the Word2vec encoding or the one-hot encoding can be used for text vectorization operation of the target breeding text to obtain target breeding text features.

[0151] Specifically, the encoding layer and the decoding layer of the self-attention mechanism or the attention neural network model such as Transformer can be used to sequentially perform position encoding and attention encoding operation on the target breeding text features to obtain target breeding text encoding; and the residual connection and linear activation operation are sequentially performed on the target breeding text encoding to obtain target text type encoding.

[0152] In detail, the target text type label is a label used to identify the text meaning of each text of the target breeding text, the feed formula text is a text recording the feed formula and feed formula ratio related information in the target breeding text, the feeding plan text is a text recording the feeding amount and feeding frequency information in the target breeding text, and the livestock product selenium content text is a text recording the content of selenium element in the livestock product prepared after each breeding period in the target breeding text.

[0153] In the embodiment of the present application, referring to Figure 3 As shown in the figure, the calculation of the selenium enrichment conversion feature according to the feed formula text, the feeding plan text and the livestock product selenium content text includes:

[0154] S31, extract feed ingredients and ingredient proportions from the feed formula text, and extract feed selenium proportions from the feed ingredients and the ingredient proportions;

[0155] S32, extract the feeding frequency and single feeding amount from the feeding plan text, and calculate the total feeding amount according to the feeding frequency and the single feeding amount;

[0156] S33, multiply the feed selenium proportion by the total feeding amount to obtain the feed selenium content;

[0157] S34, divide the selenium content of the livestock product by the feed selenium content to obtain the selenium enrichment conversion rate, and vectorize the selenium enrichment conversion rate into a selenium enrichment conversion feature.

[0158] In detail, the extraction of the feed selenium proportion from the feed ingredients and the ingredient proportion means that the selenium element proportion of each feed ingredient is matched, the selenium element proportion and the ingredient proportion are used to calculate the selenium element proportion of the whole feed, and the selenium element proportion of the whole feed is taken as the feed selenium proportion. The total feeding amount calculated according to the feeding frequency is that the total feeding frequency is calculated according to the feeding frequency, and the sum of all single feeding amounts corresponding to the total feeding frequency is taken as the total feeding amount.

[0159] In the embodiment of the application, by sequentially performing text feature extraction, feature conversion and feature segmentation on the standard breeding text set, a feed formula feature set, a feeding plan feature set and a selenium enrichment conversion feature set are obtained, so that the overall scheme of feed ratio feeding in each breeding cycle and the final selenium element biological conversion rate corresponding to each overall scheme can be determined.

[0160] S4, training a preset selenium enrichment feeding model into a selenium enrichment conversion model by using the livestock state feature set, the feed formula feature set, the feeding plan feature set and the selenium enrichment conversion feature set.

[0161] In the embodiment of the application, the selenium enrichment feeding model can be a self-attention model (Self-Attention Model) or a multiple linear regression equation (Multiple Linear Regression Equation), wherein the self-attention model is a model for processing sequence data, and the multiple linear regression equation is a regression model for establishing a linear relationship between multiple independent variables and a dependent variable.

[0162] In the embodiment of the application, the training of the preset selenium enrichment feeding model into the selenium enrichment conversion model by using the livestock state feature set, the feed formula feature set, the feeding plan feature set and the selenium enrichment conversion feature set comprises:

[0163] Fusing the livestock state feature set, the feed formula feature set and the feeding plan feature set into a livestock product preparation feature group set;

[0164] calculate the predicted conversion feature set corresponding to the livestock product preparation feature group set by using the preset selenium-rich feeding model;

[0165] calculate the loss value between the predicted conversion feature set and the selenium-rich conversion feature set;

[0166] determine whether the loss value is greater than a preset loss threshold value;

[0167] If yes, update the model parameters of the selenium-rich feeding model according to the loss value, and return to the step of calculating the predicted conversion feature set corresponding to the livestock product preparation feature group set by using the preset selenium-rich feeding model;

[0168] If no, use the updated selenium-rich feeding model as a selenium-rich conversion model.

[0169] In detail, the method of fusing the livestock state feature set, the feed formula feature set, and the feeding plan feature set into the livestock product preparation feature group set is consistent with the method of fusing the livestock type feature set, the livestock size feature set, and the livestock stage feature set into the livestock state feature set in the above step S2, which will not be repeated here.

[0170] In detail, the loss value between the predicted conversion feature set and the selenium-rich conversion feature set can be calculated by using a cross-entropy loss function or a mean square error loss function, and the model parameters of the selenium-rich feeding model can be updated according to the loss value by using a stochastic gradient descent algorithm or a fast gradient descent algorithm.

[0171] In the embodiment of the present application, by using the livestock state feature set, the feed formula feature set, the feeding plan feature set, and the selenium-rich conversion feature set, the preset selenium-rich feeding model is trained into a selenium-rich conversion model, so that a relationship model between the influencing factors such as livestock state, feed formula, and feeding plan in each breeding cycle and the selenium element biological conversion rate of the prepared selenium-rich livestock product can be established, and then subsequent targeted improvement of the breeding process and improvement of the biological conversion rate can be facilitated.

[0172] S5, acquiring real-time livestock video, extracting real-time livestock state features from the real-time livestock video, analyzing a real-time feeding scheme corresponding to the real-time livestock state features by using the selenium-rich conversion model, and preparing a selenium-rich livestock product according to the real-time feeding scheme.

[0173] In the embodiment of the present application, the real-time livestock video refers to a video obtained by real-time behavior monitoring of livestock that needs to be prepared for selenium-rich livestock product preparation. By acquiring the real-time livestock video, real-time state information of the livestock can be conveniently extracted, and then the feeding scheme can be recommended.

[0174] In the embodiment of the present application, the method for extracting real-time livestock state features from the real-time livestock video is consistent with the method for extracting livestock species feature set, livestock size feature set and livestock stage feature set from the livestock image group set in step S2, and details are not repeated here.

[0175] In the embodiment of the present application, the real-time feeding scheme corresponding to the real-time livestock state features is analyzed by using the selenium-rich conversion model, comprising:

[0176] The real-time preparation feature group set corresponding to the real-time state features is generated by using a preset random number algorithm;

[0177] The real-time conversion feature set corresponding to the real-time preparation feature group set is calculated by using the selenium-rich conversion model;

[0178] The standard conversion feature is extracted from the real-time conversion feature set by using a simulated annealing algorithm;

[0179] The real-time preparation feature group corresponding to the standard conversion feature is selected from the real-time preparation feature group set as a target preparation feature group;

[0180] The target feed formula feature and the target feeding plan feature are extracted from the target preparation feature group;

[0181] The real-time feeding scheme is generated according to the target feed formula feature and the target feeding plan feature.

[0182] Specifically, the random number algorithm can be a Pseudorandom Number Generator (PRNG) or a Gaussian distribution random number generator, the simulated annealing algorithm is a heuristic optimization algorithm used to find a global optimal solution or an approximate solution close to the optimal solution in a complex search space, and the real-time feeding scheme generated according to the target feed formula feature and the target feeding plan feature refers to preparing feed according to the feed formula corresponding to the target feed formula feature and the feeding plan corresponding to the target feeding plan feature as a real-time feeding plan.

[0183] In the embodiment of the present application, by acquiring real-time livestock video, extracting real-time livestock state features from the real-time livestock video, using the selenium-rich conversion model to analyze the real-time feeding scheme corresponding to the real-time livestock state features, and preparing the selenium-rich livestock product according to the real-time feeding scheme, the feeding scheme with the maximum selenium element biological conversion rate can be determined according to the different states and different physiological characteristics of the livestock, the flexibility of the preparation of the selenium-rich livestock product is improved, and the selenium element biological conversion rate is improved.

[0184] In the embodiment of the present application, the selenium-rich breeding data is split into period breeding data sets according to a breeding period, the period breeding data sets are cleaned into standard breeding data sets, standard breeding video sets and corresponding standard breeding text sets are extracted from the standard breeding data sets, the accuracy of the data set used for model training can be improved, and then the calculation accuracy of the subsequent model is improved, key frames are extracted from the standard breeding video sets to obtain a breeding livestock image group set, livestock species feature sets, livestock size feature sets and livestock stage feature sets are extracted from the breeding livestock image group set respectively, the livestock species feature sets, the livestock size feature sets and the livestock stage feature sets are fused into livestock state feature sets, the state features of the livestock corresponding to each breeding period can be extracted from the breeding video, and then different feeding plans can be implemented for livestock in different states, so that the biological conversion rate of the selenium-rich livestock product is improved, feed formula feature sets, feeding plan feature sets and selenium-rich conversion feature sets are obtained by sequentially performing text feature extraction, feature transcoding and feature segmentation operations on the standard breeding text sets, and the overall scheme of the feed ratio feeding in each breeding period and the final selenium element biological conversion rate corresponding to each overall scheme can be determined.

[0185] By using the livestock state feature sets, the feed formula feature sets, the feeding plan feature sets and the selenium-rich conversion feature sets to train a preset selenium-rich feeding model into a selenium-rich conversion model, a relationship model between the influencing factors such as the state of the livestock, the feed formula, the feeding plan and the selenium element biological conversion rate of the prepared selenium-rich livestock product in each breeding period can be established, and then the breeding process can be improved for subsequent targeted improvement, the biological conversion rate is improved, real-time livestock video is acquired, real-time livestock state features are extracted from the real-time livestock video, the real-time feeding scheme corresponding to the real-time livestock state features is analyzed by using the selenium-rich conversion model, and the selenium-rich livestock product is prepared according to the real-time feeding scheme. Therefore, the feeding scheme with the maximum selenium element biological conversion rate can be determined according to the different states and different physiological characteristics of the livestock, the flexibility of the preparation of the selenium-rich livestock product is improved, and the selenium element biological conversion rate is improved. Therefore, the preparation method for preparing the selenium-rich livestock product can solve the problem of low biological conversion rate during the preparation of the selenium-rich livestock product.

[0186] AsFigure 4 Figure 19 shows a functional module diagram of a preparation system for realizing selenium-rich livestock products according to an embodiment of the present application.

[0187] The preparation system 100 for realizing selenium-rich livestock products according to the present application can be installed in an electronic device. According to the functions implemented, the preparation system 100 for realizing selenium-rich livestock products can include a data splitting module 101, a video feature extraction module 102, a text feature extraction module 103, a model training module 104, and a product preparation module 105. The modules according to the present application can also be referred to as units, which refer to a series of computer program segments that can be executed by an electronic device processor and can complete a fixed function, which are stored in the memory of the electronic device.

[0188] In the present embodiment, the functions of each module / unit are as follows:

[0189] The data splitting module 101 is configured to split selenium-rich breeding data into a period breeding data set according to a breeding period, clean the period breeding data set into a standard breeding data set, extract a standard breeding video set and a corresponding standard breeding text set from the standard breeding data set.

[0190] The video feature extraction module 102 is configured to extract key frames from the standard breeding video set to obtain a breeding livestock image group set, extract a livestock species feature set, a livestock size feature set, and a livestock stage feature set from the breeding livestock image group set, respectively, and fuse the livestock species feature set, the livestock size feature set, and the livestock stage feature set into a livestock state feature set. The extraction of key frames from the standard breeding video set to obtain a breeding livestock image group set includes: selecting a standard breeding video in the standard breeding video set as a target standard breeding video one by one, and cutting the target standard breeding video into a standard breeding image set using a pre-set video segmentation tool; selecting two consecutive standard breeding images in the standard breeding image set as a target comparison breeding image group one by one, and calculating the change rate corresponding to the target comparison breeding image group using the following change rate algorithm:

[0191]

[0192] wherein S refers to the change rate, m refers to the pixel length of each standard breeding image in the target comparison breeding image group, n refers to the pixel width of each standard breeding image in the target comparison breeding image group, a is a pre-set first change weight coefficient, β is a pre-set second change weight coefficient, A max refers to the maximum gray value of the pixels in the first standard breeding image in the target comparison breeding image group, A min refers to the minimum gray value of the pixels in the first standard breeding image in the target comparison breeding image group, Bmax is the maximum gray value of the pixel in the second standard breeding picture in the target contrast breeding picture set, B min is the minimum gray value of the pixel in the second standard breeding picture in the target contrast breeding picture set, i is the i-th, and j is the j-th, A i,j is the gray value of the pixel with the pixel coordinate (i, j) in the first standard breeding picture in the target contrast breeding picture set, B i,j is the gray value of the pixel with the pixel coordinate (i, j) in the second standard breeding picture in the target contrast breeding picture set; a livestock confidence set of the variable breeding picture set is calculated by using a target detection method according to the variable rate; a breeding livestock picture set is screened from the variable breeding picture set according to the livestock confidence set, and all the breeding livestock picture sets are collected into a breeding livestock picture set set;

[0193] The text feature extraction module 103 is configured to perform text feature extraction, feature conversion and feature segmentation on the standard breeding text set in sequence to obtain a feed formula feature set, a feeding plan feature set and a selenium enrichment conversion feature set.

[0194] The model training module 104 is configured to train a preset selenium enrichment feeding model into a selenium enrichment conversion model by using the livestock state feature set, the feed formula feature set, the feeding plan feature set and the selenium enrichment conversion feature set.

[0195] The product preparation module 105 is configured to obtain a real-time livestock video, extract real-time livestock state features from the real-time livestock video, analyze a real-time feeding scheme corresponding to the real-time livestock state features by using the selenium enrichment conversion model, and prepare a selenium enrichment livestock product according to the real-time feeding scheme.

[0196] In detail, the modules in the system 100 for preparing a selenium enrichment livestock product in the embodiments of the present application are used in the same way as the technical means of the method for preparing a selenium enrichment livestock product in the above Figures 1 to 3 , and can produce the same technical effects, which will not be described here.

[0197] In the several embodiments of the present application, it should be understood that the disclosed device, system and method can be implemented in other ways. For example, the above-described system embodiments are merely illustrative, for example, the division of the modules is only a logical function division, and actual implementation can have another division manner.

[0198] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical units, i.e., may be located in one place, or may be distributed to multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0199] In addition, each functional module in various embodiments of the application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of hardware plus software functional modules.

[0200] It is obvious to those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and the present application can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application.

[0201] The embodiments of the present application can acquire and process related data based on artificial intelligence technology. Among them, artificial intelligence (AI) is to use digital computers or computer-controlled machines to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0202] In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. The plurality of units or systems stated in the system embodiment can also be realized by one unit or system through software or hardware. The words first, second, etc. are used to indicate names, not any specific order.

[0203] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit it, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. A method for preparing selenium-enriched livestock products, characterized in that, The method includes: S1: The selenium-enriched aquaculture data is split into a cycle aquaculture dataset according to the aquaculture cycle. The cycle aquaculture dataset is cleaned into a standard aquaculture dataset. The standard aquaculture video set and the corresponding standard aquaculture text set are extracted from the standard aquaculture dataset. S2: Keyframe extraction is performed on the standard livestock farming video set to obtain a livestock image set. Livestock species feature set, livestock size feature set, and livestock stage feature set are extracted from the livestock image set. These three feature sets are then merged into a livestock state feature set. The keyframe extraction from the standard livestock farming video set to obtain the livestock image set includes: S21: Select standard aquaculture videos from the standard aquaculture video set one by one as target standard aquaculture videos, and use a preset video segmentation tool to cut the target standard aquaculture videos into standard aquaculture image sets; S22: Select two consecutive standard aquaculture images from the standard aquaculture image set to form a target comparison aquaculture image set, and calculate the rate of change corresponding to the target comparison aquaculture image set using the following rate of change algorithm: ； in, This refers to the rate of change. This refers to the pixel length of each standard aquaculture image in the target comparison aquaculture image set. This refers to the pixel width of each standard aquaculture image in the target comparison aquaculture image set. It is the preset first variable weighting coefficient. It is the preset second variable weighting coefficient. This refers to the maximum grayscale value of a pixel in the first standard aquaculture image in the target comparison aquaculture image set. This refers to the minimum grayscale value of a pixel in the first standard aquaculture image in the target comparison aquaculture image set. This refers to the maximum grayscale value of a pixel in the second standard aquaculture image within the target comparison aquaculture image set. This refers to the minimum grayscale value of a pixel in the second standard aquaculture image in the target comparison aquaculture image set. It refers to the first indivual, It refers to the first indivual, This refers to the pixel coordinates of the first standard aquaculture image in the target comparison aquaculture image set. The grayscale value of the pixel, This refers to the pixel coordinates of the second standard aquaculture image in the target comparison aquaculture image set. The grayscale value of the pixel; S23: Select a variable livestock atlas from the standard livestock atlas based on the change rate, and calculate the livestock confidence set of the variable livestock atlas using a target detection method; S24: Based on the livestock confidence set, select livestock map groups from the variable livestock map set, and compile all livestock map groups into a livestock map group set; S3: Perform text feature extraction, feature transcoding, and feature segmentation operations on the standard aquaculture text set in sequence to obtain feed formula feature set, feeding plan feature set, and selenium enrichment conversion feature set; S4: Using the livestock state feature set, the feed formulation feature set, the feeding plan feature set, and the selenium enrichment conversion feature set, the preset selenium enrichment feeding model is trained into a selenium enrichment conversion model, including: The livestock status feature set, the feed formulation feature set, and the feeding plan feature set are merged into a livestock product preparation feature set; The predicted conversion feature set corresponding to the livestock product preparation feature set is calculated using a preset selenium-enriched feeding model; Calculate the loss value between the predicted transformation feature set and the selenium-enriched transformation feature set; Determine whether the loss value is greater than a preset loss threshold; If so, the model parameters of the selenium-enriched feeding model are updated according to the loss value, and the step of calculating the predicted conversion feature set corresponding to the livestock product preparation feature set using the preset selenium-enriched feeding model is returned. If not, the updated selenium-enriched feeding model will be used as the selenium-enriched conversion model. The selenium-enriched feeding model is either a self-attention model or a multiple linear regression equation. S5: Acquire real-time livestock videos, extract real-time livestock status features from the real-time livestock videos, analyze the real-time feeding plan corresponding to the real-time livestock status features using the selenium enrichment conversion model, and prepare selenium-enriched livestock products according to the real-time feeding plan.

2. The method for preparing selenium-enriched livestock products as described in claim 1, characterized in that, The step of cleaning the periodic aquaculture dataset into a standard aquaculture dataset includes: Repeated and default breeding data are sequentially filtered out from the cycle breeding dataset to obtain the initial breeding dataset; The initial aquaculture dataset was split into a text aquaculture dataset and a video aquaculture dataset according to the data format. The text aquaculture dataset is then subjected to unit normalization and outlier removal operations on each text aquaculture data in the text aquaculture dataset to obtain a standard text aquaculture dataset. The video aquaculture dataset is transcoded to obtain a standard video aquaculture dataset. The standard text aquaculture dataset and the standard video aquaculture dataset are combined into a standard aquaculture dataset.

3. The method for preparing selenium-enriched livestock products as described in claim 1, characterized in that, The method of calculating the livestock confidence set of the variable livestock atlas using object detection includes: Select one by one the variable aquaculture images in the variable aquaculture image set as the target variable aquaculture images, extract the variable aquaculture outline from the target variable aquaculture images, and generate the outline bounding box based on the variable aquaculture outline. The target aquaculture texture image is obtained by filtering the target aquaculture image using the following texture feature extraction algorithm; ； in, This refers to the coordinates in the target aquaculture texture image. The grayscale value of the pixel, It is an exponential function. It is the x-coordinate of the pixel in the target aquaculture image. It is the ordinate of the pixel in the target aquaculture image. Let be the standard deviation of the Gaussian function. It is the symbol for an imaginary number. It is pi. It is a frequency parameter. It is the x-coordinate of the pixel in the target aquaculture texture image. It is a phase parameter; Perform multi-level convolution and multi-level pooling operations on the target aquaculture texture image to obtain the target aquaculture features; The predicted bounding boxes and livestock bounding box probabilities of the target breeding features are calculated using a pre-trained livestock target detection model. The confidence scores of livestock are calculated based on the predicted bounding box, the outline bounding box, and the probability of the livestock boxes using the following confidence algorithm, and all livestock confidence scores are aggregated into a livestock confidence set: ； in, This refers to the confidence level of the livestock. This refers to the probability of the livestock frame. It is the predicted bounding box, It is the outline bounding box, It is the intersection symbol. It is the union symbol.

4. The method for preparing selenium-enriched livestock products as described in claim 1, characterized in that, The extraction of livestock species feature sets, livestock size feature sets, and livestock stage feature sets from the livestock image set includes: Select each livestock image group in the livestock image set as the target livestock image group, and extract livestock outline groups from the target livestock image groups: Based on the livestock outline group, the target livestock image group is divided into a livestock image group and a background image group; The background image group is subjected to reference object identification to obtain a reference object image group. The size characteristics of the livestock are calculated based on the reference object image group and the livestock image group. The livestock texture features, livestock color features, and livestock joint features are extracted sequentially from the livestock image group, and the livestock texture features, livestock color features, and livestock joint features are combined into livestock species features; The abdominal features and posture features of the livestock are extracted sequentially from the livestock image set, and the abdominal features and posture features of the livestock are combined into livestock stage features; All livestock size characteristics are compiled into a livestock size characteristic set, all livestock species characteristics are compiled into a livestock species characteristic set, and all livestock stage characteristics are compiled into a livestock stage characteristic set.

5. The method for preparing selenium-enriched livestock products as described in claim 1, characterized in that, The process of fusing the livestock species feature set, the livestock size feature set, and the livestock stage feature set into a livestock state feature set includes: Select livestock species features from the livestock species feature set one by one as target livestock species features, and filter livestock size features corresponding to the target livestock species features from the livestock size feature set as target livestock size features; The livestock stage features corresponding to the target livestock species features in the livestock stage feature set are taken as the target livestock stage features. The target livestock species feature is globally pooled into a dimensionality-reduced livestock species feature, the target livestock size feature is globally pooled into a dimensionality-reduced livestock size feature, and the target livestock stage feature is globally pooled into a dimensionality-reduced livestock stage feature. The reduced-dimensional livestock species features, reduced-dimensional livestock size features, and reduced-dimensional livestock stage features are fused as dimensional features into livestock state features, and all livestock state features are aggregated into a livestock state feature set.

6. The method for preparing selenium-enriched livestock products as described in claim 1, characterized in that, The process of sequentially performing text feature extraction, feature transcoding, and feature segmentation on the standard aquaculture text set yields a feed formulation feature set, a feeding plan feature set, and a selenium enrichment conversion feature set, including: Each standard aquaculture text in the standard aquaculture text set is selected as the target aquaculture text. Then, text segmentation, stop word filtering, and text vectorization are performed on the target aquaculture text in sequence to obtain the target aquaculture text features. The target aquaculture text features are sequentially subjected to positional encoding and attention encoding operations to obtain the target aquaculture text encoding; The target aquaculture text encoding is sequentially subjected to residual connection and linear activation operations to obtain the target text type encoding; Decode the target text type annotation from the target text type encoding, and split the target aquaculture text into feed formula text, feeding plan text, and livestock product selenium content text according to the target text type annotation; The feed formulation features are extracted from the feed formulation text, the feeding plan features are extracted from the feeding plan text, and the selenium enrichment conversion features are calculated based on the feed formulation text, the feeding plan text, and the selenium content text of the livestock products. All feed formulation characteristics are compiled into a feed formulation characteristic set, all feeding plan characteristics are compiled into a feeding plan characteristic set, and all selenium enrichment conversion characteristics are compiled into a selenium enrichment conversion characteristic set.

7. The method for preparing selenium-enriched livestock products as described in claim 6, characterized in that, The calculation of selenium enrichment conversion characteristics based on the feed formulation text, the feeding plan text, and the selenium content text of the livestock products includes: The feed ingredients and their proportions are extracted from the feed formula text, and the feed selenium proportion is extracted from the feed ingredients and their proportions. The feeding frequency and single feeding amount are extracted from the feeding plan text, and the total feeding amount is calculated based on the feeding frequency and the single feeding amount; Multiply the feed selenium percentage by the total feed amount to obtain the feed selenium content; Divide the selenium content of the livestock product by the selenium content of the feed to obtain the selenium enrichment conversion rate, and then vectorize the selenium enrichment conversion rate into selenium enrichment conversion characteristics.

8. The method for preparing selenium-enriched livestock products as described in claim 1, characterized in that, The process of analyzing the real-time feeding plan corresponding to the real-time livestock status characteristics using the selenium-enriched conversion model includes: A real-time feature set corresponding to the real-time livestock state features is generated using a preset random number algorithm; The real-time conversion feature set corresponding to the real-time preparation feature set is calculated using the selenium-enriched conversion model. Standard transformation features are extracted from the real-time transformation feature set using the simulated annealing algorithm; Select the real-time preparation feature group corresponding to the standard conversion feature from the real-time preparation feature group set as the target preparation feature group; Extract the target feed formulation features and target feeding plan features from the target preparation feature set; A real-time feeding plan is generated based on the target feed formulation characteristics and the target feeding plan characteristics.

9. A system for preparing selenium-enriched livestock products, characterized in that, The system includes: The data splitting module is used to split the selenium-enriched aquaculture data into a cycle aquaculture dataset according to the aquaculture cycle, clean the cycle aquaculture dataset into a standard aquaculture dataset, and extract a standard aquaculture video set and a corresponding standard aquaculture text set from the standard aquaculture dataset. The video feature extraction module is used to extract keyframes from the standard livestock farming video set to obtain a livestock image set. It extracts livestock species feature sets, livestock size feature sets, and livestock stage feature sets from the livestock image set, and merges these feature sets into a livestock state feature set. The keyframe extraction from the standard livestock farming video set to obtain the livestock image set includes: selecting standard farming videos from the standard farming video set as target standard farming videos, and using a preset video segmentation tool to cut the target standard farming videos into standard farming image sets; selecting two consecutive standard farming images from the standard farming image set to form a target comparison farming image set, and calculating the change rate corresponding to the target comparison farming image set using the following change rate algorithm: ； in, This refers to the rate of change. This refers to the pixel length of each standard aquaculture image in the target comparison aquaculture image set. This refers to the pixel width of each standard aquaculture image in the target comparison aquaculture image set. It is the preset first variable weighting coefficient. It is the preset second variable weighting coefficient. This refers to the maximum grayscale value of a pixel in the first standard aquaculture image in the target comparison aquaculture image set. This refers to the minimum grayscale value of a pixel in the first standard aquaculture image in the target comparison aquaculture image set. This refers to the maximum grayscale value of a pixel in the second standard aquaculture image within the target comparison aquaculture image set. This refers to the minimum grayscale value of a pixel in the second standard aquaculture image in the target comparison aquaculture image set. It refers to the first indivual, It refers to the first indivual, This refers to the pixel coordinates of the first standard aquaculture image in the target comparison aquaculture image set. The grayscale value of the pixel, This refers to the pixel coordinates of the second standard aquaculture image in the target comparison aquaculture image set. The grayscale value of the pixel; the variable livestock map set is selected from the standard livestock map set according to the change rate, and the livestock confidence set of the variable livestock map set is calculated using the target detection method; the livestock map group is selected from the variable livestock map set according to the livestock confidence set, and all the livestock map groups are combined into a livestock map group set. The text feature extraction module is used to sequentially perform text feature extraction, feature transcoding, and feature segmentation operations on the standard aquaculture text set to obtain feed formula feature set, feeding plan feature set, and selenium enrichment conversion feature set. The model training module is used to train a preset selenium-enriched feeding model into a selenium-enriched conversion model using the livestock state feature set, the feed formulation feature set, the feeding plan feature set, and the selenium-enriched conversion feature set. This includes: fusing the livestock state feature set, the feed formulation feature set, and the feeding plan feature set into a livestock product preparation feature set; calculating the predicted conversion feature set corresponding to the livestock product preparation feature set using the preset selenium-enriched feeding model; calculating the loss value between the predicted conversion feature set and the selenium-enriched conversion feature set; determining whether the loss value is greater than a preset loss threshold; if so, updating the model parameters of the selenium-enriched feeding model based on the loss value and returning to the step of calculating the predicted conversion feature set corresponding to the livestock product preparation feature set using the preset selenium-enriched feeding model; if not, using the updated selenium-enriched feeding model as the selenium-enriched conversion model; wherein the selenium-enriched feeding model is a self-attention model or a multiple linear regression equation. The product preparation module is used to acquire real-time livestock videos, extract real-time livestock status features from the real-time livestock videos, analyze the real-time feeding plan corresponding to the real-time livestock status features using the selenium enrichment conversion model, and prepare selenium-enriched livestock products according to the real-time feeding plan.

Images