A method and device for predicting selenium deposition in pigs

Abstract

The present application provides a method and device for predicting the amount of selenium deposited in a pig, and relates to the field of data processing technology. The method comprises: obtaining target images from multiple camera devices, environmental data of a pig house, and feeding data, wherein the target image is used to indicate at least one target object representing a pig; performing image processing on the target image to obtain image features and temporal and spatial features of the target object contained in the target image, wherein the temporal and spatial features are used to indicate the position and appearance time of the target object in the target image; determining the motion trajectory of each target object based on the image features and temporal and spatial features; inputting the environmental data of the target object at any time, the amount of motion data determined by the motion trajectory, and the feeding data into a selenium deposition prediction model, performing prediction, and outputting the amount of selenium deposited by the target object when the planned feeding time is reached. The present application can improve the accuracy of predicting the amount of selenium deposited in a pig.

Description

Technical Field

[0001] The invention relates to the technical field of data processing, in particular to a method and a device for predicting selenium deposition in a pig. Background Art

[0002] Selenium is an essential trace element that participates in many important physiological functions, such as anti-oxidation, disease prevention and treatment, and promoting animal growth. As an important economic animal, the selenium content in pigs not only directly affects growth performance and immune function, but is also closely related to meat quality and human health. Therefore, accurately predicting selenium deposition in pigs has important theoretical and application value, thereby optimizing the amount of selenium added in feed and avoiding health problems and economic losses caused by selenium deficiency or excess.

[0003] For the prediction of selenium deposition in pigs, existing methods mainly use pig breed, feeding cycle, and feed selenium content as input data, extract input data features with pig nutrition as constraints, and construct a nonlinear relationship between features and selenium content through learning models, and finally obtain the prediction results of selenium content in pigs. However, this type of analysis method often ignores the differences in individual pig movement and environmental data. In the actual breeding process, there are significant differences in the exercise conditions (such as exercise volume) and environmental conditions such as temperature and humidity of different pigs, all of which have an important impact on the deposition of selenium in pigs. Specifically, in terms of exercise, if pigs do not exercise enough for a long time, it will lead to a decrease in metabolic level and slow blood circulation, thereby inhibiting the absorption, distribution and metabolic efficiency of selenium. In this case, the amount of selenium deposited in muscle tissue may decrease, and more selenium may be excreted from the body in an underutilized form; if pigs exercise excessively, it will cause the body to produce a large amount of reactive oxygen species (ROS), leading to oxidative stress, thereby significantly increasing the body's demand for selenium. Temperature and humidity, as key environmental factors, act together on the growth environment of pigs, affecting their feed intake, metabolism and health status. The appropriate combination of temperature and humidity can optimize the growth environment of pigs, improve their absorption and utilization efficiency of selenium, and thus promote the deposition of selenium in the body. Studies have shown that under appropriate temperature and humidity conditions, the growth performance and selenium deposition efficiency of pigs are significantly improved. In short, these methods often ignore the differences in individual pig movement and environmental data, resulting in low accuracy in predicting the amount of selenium deposition in pigs in practical applications. Summary of the invention

[0004] In order to solve the above-mentioned problems in the prior art, the present invention provides a method and device for predicting the amount of selenium deposited in a pig, which can solve the problem of low accuracy in predicting the amount of selenium deposited in a pig in the prior art.

[0005] In a first aspect, the present application provides a method for predicting selenium deposition in a pig, comprising:

[0006] Acquire target images from a plurality of camera devices, environmental data of a piggery, and feeding data, wherein the target images are used to indicate at least one target object representing a pig, and the feeding data includes a planned feeding time;

[0007] Performing image processing on the target image to obtain image features and spatiotemporal features of the target object contained in the target image, wherein the spatiotemporal features are used to indicate the position and appearance time of the target object in the target image;

[0008] Determine the motion trajectory of each target object based on the image features and the spatiotemporal features;

[0009] The environmental data of the target object at any time, the exercise data determined by the motion trajectory, and the feeding data are input into a pre-established selenium deposition prediction model to perform prediction, and the selenium deposition amount of the target object when the planned feeding time is reached is output.

[0010] In a possible implementation, performing image processing on the target image to obtain image features and temporal and spatial features of the target object contained in the target image includes:

[0011] For any target image, target detection is performed on the target image to obtain at least one target detection frame, wherein the target image carries the device identification and acquisition time of the camera device to which it belongs, and the target detection frame is used to indicate any target object on the target image; based on the target detection frame, the target image is divided into sub-image blocks corresponding to each target object; feature extraction is performed on the sub-image blocks to obtain a feature map, and bilinear interpolation processing is performed on the feature map to obtain a re-identification feature as the image feature of the target object; based on the frame position data of the target detection frame on the target image and the acquisition time and device identification of the target image, the spatiotemporal characteristics of the target object indicated by the target detection frame are determined.

[0012] In a possible implementation, determining the motion trajectory of each target object based on the image feature and the spatiotemporal feature includes:

[0013] The image features and spatiotemporal features of the target object contained in the target image are fused to obtain the fused features of the target image; for any target object, all target images are traversed, and the detection results of the target object are matched based on the fused features, wherein the detection results are used to indicate the target images capturing the target object at multiple time moments and the target detection frames identifying the target object on the target images; based on the frame position data of the target detection frames indicated by the detection results, the target position data of the target object in the target image to which it belongs at each time moment is determined; based on the device identification corresponding to the target image indicated by the detection results, the device position data of the corresponding camera device is determined; based on the target position data and the device position data, the cumulative moving distance of the target object at each time moment is determined, and the motion trajectory of the target object is generated according to the cumulative moving distance.

[0014] In a possible implementation, for any target object, traversing all target images and matching the detection result of the target object based on the fusion feature includes:

[0015] The target image at a specified moment is divided into a first image, and the target images at other moments except the specified moment are divided into a second image; based on the fusion feature of the first image, a first association feature of the first image is determined, and based on the fusion feature of the second image, a second association feature of the second image is determined, wherein the first association feature is used to indicate the association degree between the target objects in the first image, and the second association feature is used to indicate the association degree between the target objects in any of the second images and the target objects in other second images; taking any target object in the first image as a detection object, matching the first association feature of the detection object with the second association feature corresponding to each target object in the second image, and determining the similarity between the detection object and the target object in the second image; screening out the target object in the second image corresponding to the similarity greater than a preset threshold as the matching object of the detection object; determining the detection result of the detection object according to the target image to which the matching object belongs and the target detection box on the target image used to indicate the matching object.

[0016] In a possible implementation, screening out the target object in the second image corresponding to the similarity being greater than a preset threshold as a matching object for the detection object includes:

[0017] When the number of screened target objects does not exceed the preset quantity threshold, the screened target objects are used as the matching objects; when the number of screened target objects exceeds the quantity threshold, the similarities of the screened target objects are sorted from large to small, and the first a target objects in the sorting are selected as the matching objects, where a is the quantity threshold.

[0018] In a possible implementation manner, determining a first associated feature of the first image based on the fusion feature of the first image, and determining a second associated feature of the second image based on the fusion feature of the second image includes:

[0019] The preset self-attention model is used to extract the fusion features of the first image to obtain the first associated features; the preset cross-attention model is used to extract the fusion features of the second image to obtain the second associated features.

[0020] In one possible implementation, the environmental data includes temperature data, humidity data and lighting data of the pig house; the feeding data also includes feeding cycle data, feed formula data and feed additive data, the feeding cycle data includes the number of feedings per day and the amount of feeding each time, and the feed formula data includes the selenium content per unit mass in the feed.

[0021] In a possible implementation, the environmental data of the target object at any time, the exercise data determined by the movement trajectory, and the feeding data are input into a pre-established selenium deposition prediction model to perform prediction, and the selenium deposition amount of the target object when the planned feeding time is reached is output, including:

[0022] Obtain a basal absorption rate, a weight coefficient, and feeding environment parameters related to the feeding process, wherein the feeding environment parameters include an environmental humidity parameter and an environmental temperature parameter, and the weight coefficient includes an exercise weight coefficient for characterizing the degree of influence of exercise on absorption efficiency, a temperature weight coefficient for characterizing the degree of influence of temperature on absorption efficiency, and a humidity weight coefficient for characterizing the degree of influence of humidity on absorption efficiency; construct an absorption rate model based on the basal absorption rate, the weight coefficient, the feeding environment data, the temperature data, the humidity data, and the amount of exercise data determined by the motion trajectory, and the amount of exercise data is used to characterize the cumulative moving distance of the target object; obtain a basal metabolic consumption parameter, an exercise saturation distance, an additive effect parameter related to the feed additive data, and a consumption coefficient characterizing the increase in selenium consumption caused by oxidative stress; determine selenium intake based on the feeding cycle data, the feed formula data, and the feed additive data; construct a metabolic model based on the basal metabolic consumption parameter, the exercise saturation distance, the additive effect parameter, the consumption coefficient, the selenium intake, and the absorption rate model; generate corresponding training labels based on the simulated deposition content output by the metabolic model; and train the selenium deposition prediction model using the training labels.

[0023] In a possible implementation, the environmental data of the target object at any time, the exercise data determined by the movement trajectory, and the feeding data are input into a pre-established selenium deposition prediction model to perform prediction, and the selenium deposition amount of the target object when the planned feeding time is reached is output, including:

[0024] The absorption rate model is determined by the following formula (1), including:

[0025] (1)

[0026] in, represents the absorptivity output by the absorptivity model, is the basic absorption rate, is the motion weight coefficient, is the exercise volume data, is the temperature weight coefficient, is the temperature data, is the ambient temperature parameter, is the humidity weight coefficient, is the humidity data, is the environmental humidity parameter;

[0027] The metabolic model is determined by the following formula (2), including:

[0028] (2)

[0029] in, represents the simulated deposition content output by the metabolic model, is the selenium intake, is the motion saturation distance, is the additive effect parameter, is the basal metabolic consumption parameter, is the consumption coefficient, is the temperature difference determined by the temperature data and the ambient temperature parameter, is a humidity difference determined by the humidity data and the ambient humidity parameter.

[0030] In a second aspect, the present application provides a device for predicting selenium deposition in a pig, comprising:

[0031] A data acquisition module, used to acquire target images from a plurality of camera devices, environmental data of a piggery, and feeding data, wherein the target images are used to indicate at least one target object representing a pig, and the feeding data includes a planned feeding time;

[0032] An image processing module, used to perform image processing on the target image to obtain image features and spatiotemporal features of the target object contained in the target image, wherein the spatiotemporal features are used to indicate the position and appearance time of the target object in the target image;

[0033] A motion analysis module, used to determine the motion trajectory of each target object based on the image features and the spatiotemporal features;

[0034] The prediction module is used to input the environmental data of the target object at any time, the exercise data determined by the motion trajectory, and the feeding data into a pre-established selenium deposition prediction model to make a prediction and output the selenium deposition amount of the target object when the planned feeding time is reached.

[0035] The method and device for predicting the amount of selenium deposits in pigs provided in the embodiments of the present application collect target images of pigs by configuring multiple cameras, and obtain image features and temporal and spatial features of the target object in the target image through image processing of the target image, thereby improving the accuracy of extracting the target object. At the same time, the motion state and motion trend of the target object are taken into account to generate the motion trajectory of the target object, thereby improving the accuracy of target tracking, and thus combining the motion data, environmental data and feeding data of the target object to predict the amount of selenium deposits of the target object when the planned feeding time is reached. Taking into account the differences in individual pig movements, this method can solve the problem of low accuracy in predicting the amount of selenium deposits in pigs in the prior art, thereby improving the accuracy of selenium deposition prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A schematic diagram of a method for predicting selenium deposition in a pig provided in an embodiment of the present application;

[0037] Figure 2 A schematic diagram of the structure of the self-attention model provided in the embodiment of the present application;

[0038] Figure 3 A schematic diagram of the structure of the cross-attention model provided in the embodiment of the present application;

[0039] Figure 4 A schematic diagram of the structure of a device for predicting selenium deposition in a pig provided in an embodiment of the present application. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] Figure 1 A schematic flow chart of a method for predicting selenium deposition in a pig provided in an embodiment of the present application, the method comprising steps S101 to S104.

[0042] S101, acquiring target images from a plurality of camera devices, environmental data of a pig house, and feeding data, wherein the target images are used to indicate at least one target object representing a pig, and the feeding data include a planned feeding time.

[0043] In some embodiments, the environmental data includes temperature data, humidity data and lighting data of the pig house; the feeding data also includes feeding cycle data, feed formula data and feed additive data, the feeding cycle data includes the number of feedings per day and the amount of feeding each time, and the feed formula data includes the selenium content per unit mass in the feed.

[0044] In this application, the feeding environment has a certain influence on the selenium deposition in the pig body. For example, environmental factors such as temperature, humidity, and light may affect the appetite and metabolic process of the pig, thereby affecting the intake and deposition of selenium in the pig body. Specifically, this application obtains environmental data, such as temperature parameters, humidity parameters, and light parameters in the pig house, by configuring sensors such as digital temperature sensors, digital humidity sensors, and digital light sensors. Therefore, the temperature data obtained from the digital temperature sensor , Represents the temperature data at time T, and obtains the humidity data from the digital humidity sensor , Represents the humidity data at time T, and obtains the light data of the digital light sensor , Represents the lighting data at time T.

[0045] It should be noted that the temperature data, humidity data and light data in the above pig house all showed an increase first and then a decrease in the effect of selenium deposition in pigs. ) may cause respiratory dryness and stress response, affect feed intake and metabolic balance, and reduce selenium absorption efficiency; suitable environment ( ) The pig's respiratory tract and skin are in good health and its metabolic function is normal, which is conducive to the efficient absorption and deposition of selenium; high humidity environment ( ) can easily lead to the growth of pathogenic microorganisms, increase the risk of disease, weaken the immune function of pigs, increase the demand for selenium, and reduce the deposition efficiency. In addition, the feeding cycle data includes the number of feedings per day, the amount of each feeding, and the planned feeding time (that is, the total planned feeding days). Feed formula data includes the selenium content per unit mass in the feed. Feed additive data indicates whether the current feed contains promoters or inhibitors (promoters such as citric acid, formic acid, lactic acid, inhibitors such as copper, zinc, and iron). Specifically, obtain feeding data , Indicates the planned feeding time (in this application set to a fixed value), Indicates the number of times of feeding per day (in this application set to a fixed value), Indicates the amount of feeding each time (in this application Set as a variable), Indicates the selenium content per unit mass in feed (in this application Set as a variable), Indicates whether the feed contains accelerators or inhibitors (e.g., 1 indicates an accelerator, 0 indicates both an accelerator and an inhibitor, and -1 indicates an inhibitor).

[0046] It should be noted that each target image contains one or more target objects, and the target objects on each target image are not completely the same. In this application, the camera equipment is set according to the specific size of the pig house to ensure that the set camera field of view can cover the entire pig house and capture the movement of the pigs in the pig house. It should be noted that the pigs in the target image are taken as the target objects, and the target image carries the device identification and acquisition time of the camera equipment to which it belongs, that is, the set of target images is represented as , The device identifier of the camera device that captured the target image. Indicates the number of camera devices. represents the time (i.e., the acquisition time of the target image), Indicates the total duration, Indicates that at time t, It should be noted that by configuring multiple camera devices, target images from multiple camera devices can be acquired at the same time. Similarly, for the same camera device, target images at multiple times can be acquired.

[0047] S102: Perform image processing on the target image to obtain image features and spatiotemporal features of the target object contained in the target image, wherein the spatiotemporal features are used to indicate the position and appearance time of the target object in the target image.

[0048] In the present application, the target image is processed to extract the image features and spatiotemporal features of the target object in the target image. Specifically, the image features may be the color, edge, texture, shape and other features of the target object in the target image, which is conducive to accurately identifying and locking the target object in a complex environment, thereby improving the accuracy of target tracking. The spatiotemporal features may be the changes in the speed, acceleration, and position of the target object over time, which can reflect the motion state and motion trend of the target, thereby improving the positioning accuracy of the target object.

[0049] S103: Determine the motion trajectory of each target object based on the image features and the spatiotemporal features.

[0050] In the present application, for each target object, the motion trajectory of the target object is generated through the image features and spatiotemporal features of the target object in one or more target images (i.e., the target image can be multiple target images acquired by different camera devices at the same time, multiple target images acquired by the same camera device at different times, and multiple target images acquired by different camera devices at different times). The motion trajectory can reflect the motion path and motion state of the target object, thereby obtaining the cumulative moving distance of the target object at a certain time, i.e., the motion amount data of the target object.

[0051] Therefore, the present application generates the motion trajectory of the target object through the image features and spatiotemporal features of the target object, thereby accurately identifying the target object and improving the accuracy of extracting motion characteristics, thereby improving the accuracy and real-time performance of the motion trajectory.

[0052] S104, inputting the environmental data of the target object at any time, the exercise data determined by the motion trajectory, and the feeding data into a pre-established selenium deposition prediction model, performing prediction, and outputting the selenium deposition amount of the target object when the planned feeding time is reached.

[0053] It should be noted that the prior art does not take into account the differences in the movement conditions of different pigs during the actual breeding process, and insufficient or excessive exercise greatly affects the amount of selenium deposited in the pig's body, resulting in inaccurate predicted selenium deposits in the pig's body. In this regard, the present application combines the environmental data at any time, the movement data of the target object (i.e., the cumulative moving distance at that time) and the current feeding data, and uses a selenium deposition prediction model to predict, and obtains the selenium deposition amount of the target object when the planned feeding time is reached. Therefore, the present application analyzes the movement of each pig, takes into account the differences in individual pig movement, and thus combines the current environmental data, feeding data and the movement of the pig to predict the amount of selenium deposited in the pig's body, thereby greatly improving the accuracy of the prediction.

[0054] In some embodiments, performing image processing on the target image to obtain image features and temporal and spatial features of the target object contained in the target image includes:

[0055] For any target image, target detection is performed on the target image to obtain at least one target detection frame, wherein the target image carries a device identification and acquisition time of a corresponding camera device, and the target detection frame is used to indicate any target object on the target image;

[0056] According to the target detection frame, the target image is divided into sub-image blocks corresponding to each target object;

[0057] Extracting features from the sub-image block to obtain a feature map, and performing bilinear interpolation processing on the feature map to obtain a re-identification feature as an image feature of the target object;

[0058] Based on the frame position data of the target detection frame on the target image and the acquisition time and device identification of the target image, the spatiotemporal characteristics of the target object indicated by the target detection frame are determined.

[0059] In this embodiment, target detection is performed on the target image to achieve classification and location of the target object in the target image. It should be noted that the target detection frame is used to mark the position of the target object in the target image. The target detection frame can be a rectangular frame, and the horizontal and vertical coordinates of the upper left corner and the lower right corner are usually used to determine the position of the target detection frame. Exemplarily, the frame position data of the target detection frame on the target image can be expressed as , Represents the horizontal coordinate of the upper left corner of the nth target detection box (corresponding to the nth target object), Represents the ordinate of the upper left corner of the nth target detection box, Represents the horizontal coordinate of the lower right corner of the nth target detection box, Indicates the ordinate of the lower right corner of the nth target detection box. Optionally, CenterNet (i.e., anchor-free target detection network) is used to perform target detection on the target image and output multiple target detection boxes that calibrate the target objects.

[0060] Furthermore, since a target detection box corresponds to a target object, the target image is divided into one or more sub-image blocks according to the target detection box. Furthermore, feature extraction is performed on the sub-image blocks containing the target object to obtain a feature map, and then the extracted feature map is scaled to a fixed size using a bilinear interpolation algorithm. , thus obtaining the re-identification feature. Specifically, for the target image The corresponding sub-image block set Perform image feature extraction to obtain a set of re-identification features , represents the re-identification feature of the nth target object (i.e., the image feature of the nth target object), A collection of image features representing the target object in the target image.

[0061] In addition, based on the frame position data of the target detection frame corresponding to the target object And the acquisition time of the target image , Equipment Identification , generate the spatiotemporal features of the target object ,Should Indicated in the target image The spatiotemporal characteristics of the nth target object.

[0062] Therefore, this embodiment performs target detection and feature extraction on the target object in the target image to determine the image features of the target object, thereby improving the accuracy of identifying the target object. In addition, the spatiotemporal features of the target object are determined through the frame position data of the target detection frame where the target object is located and the device identification and acquisition time of the target image, which can reflect the motion state and motion trend of the target object and help improve the accuracy of locating the target object.

[0063] In some embodiments, step 103 includes:

[0064] Fusing the image features and the temporal and spatial features of the target object contained in the target image to obtain fused features of the target image;

[0065] For any target object, traverse all target images, and match the detection result of the target object based on the fusion feature, wherein the detection result is used to indicate the target image capturing the target object at multiple moments and the target detection frame identifying the target object on the target image;

[0066] Determining target position data of the target object in the target image at each moment based on the frame position data of the target detection frame indicated by the detection result;

[0067] Determine device location data of a corresponding camera device based on a device identifier corresponding to the target image indicated by the detection result;

[0068] The cumulative moving distance of the target object at each moment is determined according to the target position data and the device position data, and the motion trajectory of the target object is generated according to the cumulative moving distance.

[0069] In this embodiment, the image features and spatiotemporal features of the target object contained in the target image are fused to obtain the fused features of the target image, so as to merge the features at different levels, thereby improving the amount of information and accuracy. , the set of image features of the target object contained in the target image A collection of spatiotemporal features Perform feature fusion to obtain the fusion features of the target image , Indicates that at time t, The fusion features corresponding to the target image acquired by the camera device. Then, based on the fusion features representing the target image in the target image, all target images are traversed to match the target objects, and the detection result of each target object (that is, all target objects in all target images) is obtained, that is, the detection result indicates the same target object on different target images. It should be noted that the detection result of each target object is used to characterize the target image that captures the target object, that is, the same target object may appear in target images captured by different cameras at the same time, target images captured by the same camera device at different times, and target images captured by different cameras at different times. At the same time, the detection result also indicates the situation of the target detection frame marking the target object on the target image, and the target position data of the target object in the target image to which it belongs is determined based on the frame position data of the target detection frame.

[0070] For example, the detection result includes that the second target object (marked as A2) at the first moment matches the first target object (marked as B1) at the second moment, then according to the target object A2 from the target image (i.e., the target image acquired by the camera device identified as 1 at the first moment), the fusion features of the target image Extract the fusion feature corresponding to the target object A2 , represents the image features of the second target object A2, Represents the spatiotemporal features of the second target object A2. Similarly, based on the target object B1 from the target image (i.e., the target image acquired by the camera device identified as 1 at the second moment), the fusion features of the target image Extract the fusion feature corresponding to the target object B1 , represents the image features of the first target object B1, represents the spatiotemporal features of the first target object B1. Thus, the frame position data in the spatiotemporal features of the target object A2 is extracted to obtain the target object A2 in the target image. The target position data in the target image is obtained based on the box position data in the spatiotemporal features of the target object B1. The target location data in .

[0071] Further, according to the device identification corresponding to the target image of the same target object, the device position data between the corresponding camera devices is determined. Then, according to the target position data and device position data of the target object, the cumulative moving distance of the target object at each moment is determined, thereby generating the motion trajectory of the target object. Exemplarily, traversing n target objects, the cumulative moving distance of the target object is obtained, which is expressed as , Represents the cumulative moving distance of the nth target object at time t.

[0072] Therefore, this embodiment matches the target object in all target images, determines the target position data where the same target object appears and the device position data of the camera device that captures the target object, thereby determining the cumulative moving distance of the target object to generate the motion trajectory of the target object, thereby estimating the amount of exercise of the pigs, and taking into account the different movement conditions of pigs, thereby improving the accuracy of selenium deposition prediction.

[0073] In some embodiments, for any target object, traversing all target images and matching the detection result of the target object based on the fusion feature includes:

[0074] Classify the target image at a specified time as a first image, and classify the target image at other times except the specified time as a second image;

[0075] Determine a first correlation feature of the first image based on the fusion feature of the first image, and determine a second correlation feature of the second image based on the fusion feature of the second image, wherein the first correlation feature is used to indicate the correlation between target objects in the first image, and the second correlation feature is used to indicate the correlation between target objects in any second image and target objects in other second images;

[0076] Taking any target object in the first image as a detection object, matching a first correlation feature for the detection object with a second correlation feature corresponding to each target object in the second image, and determining a similarity between the detection object and the target object in the second image;

[0077] Screening out the target object in the second image corresponding to the similarity greater than a preset threshold as a matching object for the detection object;

[0078] A detection result of the detection object is determined according to a target image to which the matching object belongs and a target detection frame on the target image used to indicate the matching object.

[0079] In this embodiment, the target image is divided by time, and a first image representing the target image at a specified time and a second image representing the target image at other times other than the specified time are obtained. It should be noted that the first image includes at least one target image captured by different camera devices at the specified time, and the second image includes at least one target image captured by different camera devices at other times other than the specified time. Exemplarily, the target image set is divided by time T as the dividing value. Divide into first image and the second image ( , , …, ),in, represents the target image captured by each camera device at the Tth time. Correspondingly, the fusion feature of the first image is expressed as , that is, the fusion features of the target images collected by each camera device at the Tth time, the fusion features of the second image are expressed as ( ).

[0080] Further, through the fusion features of the first image, the dependency between the target objects in the first image, that is, the relationship between the target objects in multiple target images at the same time, is extracted to generate a first association feature. Then, through the fusion features of the second image, the association between the target objects in any second image and the target objects in other second images, that is, the relationship between the target objects in multiple target images at each time except the specified time and the target objects in multiple target images at other times, is generated to generate a second association feature. Further, taking any target object in the first image as the detection object, based on the first association feature and the second association feature, the matching object corresponding to the detection object is matched in the second image. Thus, according to the second image to which the matching object belongs and the target detection frame marking the matching object in the second image, the device identification, acquisition time and frame position data of the target detection frame corresponding to the second image are extracted to generate the detection result of the detection object. Wherein, the detection result includes at least one matching object and the target image and target detection frame corresponding to the matching object.

[0081] For example, represents the first associated feature, Represents the second association feature, through the formula Determine the similarity, where The matrix representing the similarity, Represents the transpose of the second associated feature. It should be noted that each element of the similarity matrix represents the similarity between each detection object and each target object in the second image. Exemplarily, the first row of the matrix represents the similarity between the first detection object in the first image and each target object in the second image, and the target object corresponding to the similarity in the first row greater than a preset threshold (such as 0.5) is screened out as the matching object of the detection object. Similarly, each detection object in the first image is traversed, one or more matching objects for each detection object are determined, and the corresponding detection results are generated.

[0082] Therefore, this embodiment fuses the image features and spatiotemporal features of the target object contained in the target image to obtain the fusion features of the target image, thereby merging features at different levels and taking into account the correlation between features at different levels. Subsequently, by determining the first correlation feature and the second correlation feature, taking into account the correlation and difference between multiple target objects in the target images captured by different cameras at the same time, and taking into account the correlation and difference between multiple target objects in the target images captured by different cameras at any time and multiple target objects in the target images captured by different cameras at other times, the information correlation is effectively extracted, thereby improving the accuracy of target object matching. Thus, the target object is matched through the first correlation feature and the second correlation feature, and the matching object and the corresponding detection result of the detection object are screened out in combination with the similarity, thereby improving the accuracy of target matching.

[0083] In some embodiments, the step of screening out the target object in the second image corresponding to the similarity being greater than a preset threshold as a matching object for the detection object includes:

[0084] When the number of screened target objects does not exceed the preset quantity threshold, the screened target objects are used as the matching objects; when the number of screened target objects exceeds the quantity threshold, the similarities of the screened target objects are sorted from large to small, and the first a target objects in the sorting are selected as the matching objects, where a is the quantity threshold.

[0085] In this embodiment, it is detected whether the number of the screened target objects exceeds a preset number threshold. If so, the similarity of the screened target objects is sorted from large to small, and the first a target objects are screened as matching objects. Otherwise, the screened target objects are used as matching objects. Optionally, the setting of a is associated with the number threshold. Exemplarily, the elements with a similarity greater than 0.5 in the first row of the similarity matrix (i.e., the target objects of the second image) are screened. If the number of elements with a similarity greater than 0.5 in the first row of the similarity matrix is ​​greater than the number threshold (such as T-1), they are sorted from large to small according to the similarity, and the target objects corresponding to the elements sorted in the first (T-1) position are used as matching objects. Therefore, this embodiment determines the matching object of the detection object by judging whether the number of the screened target objects exceeds the number threshold and selecting the target objects with higher similarity, thereby improving the accuracy of the target object matching.

[0086] In some embodiments, determining the first associated feature of the first image based on the fusion feature of the first image, and determining the second associated feature of the second image based on the fusion feature of the second image includes:

[0087] The preset self-attention model is used to extract the fusion features of the first image to obtain the first associated features; the preset cross-attention model is used to extract the fusion features of the second image to obtain the second associated features.

[0088] For example, Figure 2 As shown, the self-attention model includes a self-attention mechanism layer 21 (Self-Attention), a residual connection and normalization layer 22 (Add&Norm), a feedforward layer 23 (Feed Forward), and a residual connection and normalization layer 24 (Add&Norm) arranged in cascade order. It should be noted that the self-attention model can focus on the relationship between the internal elements of a single input sequence, so the fusion features of the first image are input into the self-attention model to extract the relationship between the target objects in the first image and output the first correlation feature. Figure 3 As shown, the cross-attention model includes a self-attention mechanism layer 31 (Self-Attention), a residual connection and normalization layer 32 (Add&Norm), a cross-attention mechanism layer 33 (Cross-Attention), a residual connection and normalization layer 34 (Add&Norm), a feed forward layer 35 (Feed Forward), and a residual connection and normalization layer 36 (Add&Norm) arranged in cascade order. It should be noted that the cross-attention model can focus on the interaction of elements between two different input sequences, so the fusion features of the second image are input into the cross-attention model to extract the relationship between each target object in any second image and each target object in other second images, and output the second correlation feature. Therefore, this embodiment extracts the first correlation feature through the self-attention model and extracts the second correlation feature through the cross-attention model, thereby improving the accuracy of data feature relationship extraction.

[0089] In some embodiments, step S104 includes:

[0090] Obtaining a basal absorption rate, a weight coefficient, and feeding environment parameters related to the feeding process, wherein the feeding environment parameters include an environmental humidity parameter and an environmental temperature parameter, and the weight coefficients include an exercise weight coefficient for characterizing the degree of influence of exercise on the absorption efficiency, a temperature weight coefficient for characterizing the degree of influence of temperature on the absorption efficiency, and a humidity weight coefficient for characterizing the degree of influence of humidity on the absorption efficiency;

[0091] constructing an absorption rate model based on the basic absorption rate, the weight coefficient, the feeding environment data, the temperature data, the humidity data, and the movement volume data determined by the movement trajectory, wherein the movement volume data is used to characterize the cumulative movement distance of the target object;

[0092] Obtaining basal metabolic consumption parameters, exercise saturation distance, additive effect parameters related to the feed additive data, and a consumption coefficient characterizing increased selenium consumption caused by oxidative stress;

[0093] determining selenium intake based on the feeding cycle data, the feed formula data and the feed additive data;

[0094] Constructing a metabolic model based on the basal metabolic consumption parameter, the exercise saturation distance, the additive effect parameter, the consumption coefficient, the selenium intake and the absorption rate model;

[0095] generating corresponding training labels according to the simulated deposition content output by the metabolic model;

[0096] The selenium deposition prediction model is trained using the training labels.

[0097] In this embodiment, the environmental humidity parameter represents the optimal humidity during the feeding process, and the environmental temperature parameter represents the optimal temperature during the feeding process. The additive effect parameter represents the additive inhibition or promotion parameter. Optionally, the absorption rate model is determined by formula (1):

[0098] (1)

[0099] in, represents the absorptivity output by the absorptivity model, is the basic absorption rate, is the motion weight coefficient, is the exercise volume data (the exercise volume data may be the accumulated moving distance when the planned feeding time is reached), is the temperature weight coefficient, is the temperature data, is the ambient temperature parameter, is the humidity weight coefficient, is the humidity data, is the environmental humidity parameter.

[0100] Further, a metabolic model is constructed based on the above absorption rate model. Optionally, the metabolic model is determined by the following formula (2), including:

[0101] (2)

[0102] in, represents the simulated deposition content output by the metabolic model, is the selenium intake, is the motion saturation distance, is the additive effect parameter, is the basal metabolic consumption parameter, is the consumption coefficient, is the temperature difference determined by the temperature data and the ambient temperature parameter (i.e. ), is the humidity difference determined by the humidity data and the ambient humidity parameter (i.e. ).

[0103] Furthermore, the training label is determined based on the simulated deposition content output by the metabolic model. In the present application, the internal parameters of the selenium deposition prediction model are adjusted by comparing the difference between its predicted output and the training label, thereby gradually optimizing its performance. Thus, the exercise data, environmental data, and feeding data at a certain moment are input into the trained selenium deposition prediction model, and the prediction is performed, and the selenium deposition in the pig body when the planned feeding time is reached is output.

[0104] Therefore, this embodiment combines environmental data, feeding data and exercise data to construct an absorption rate model, and then combines the absorption rate, environmental data, feeding data and exercise data output by the absorption rate model to construct a metabolic model, thereby generating training labels based on the simulated deposition content output by the metabolic model, and thereby training a selenium deposition prediction model, taking into account the impact of environmental, exercise and feeding factors on selenium deposition prediction, thereby improving the accuracy of the model and thus improving the prediction accuracy.

[0105] Figure 4 The schematic diagram of the structure of a device for predicting selenium deposition in pigs provided in an embodiment of the present application, wherein the device 400 for predicting selenium deposition in pigs comprises:

[0106] A data acquisition module 401 is used to acquire target images from multiple camera devices, environmental data of the pig house, and feeding data, wherein the target image is used to indicate at least one target object representing the pigs, and the feeding data includes a planned feeding time;

[0107] An image processing module 402 is used to perform image processing on the target image to obtain image features and spatiotemporal features of the target object contained in the target image, wherein the spatiotemporal features are used to indicate the position and appearance time of the target object in the target image;

[0108] A motion analysis module 403, used to determine the motion trajectory of each target object based on the image features and the spatiotemporal features;

[0109] The prediction module 404 is used to input the environmental data of the target object at any time, the exercise data determined by the motion trajectory, and the feeding data into a pre-established selenium deposition prediction model to make a prediction and output the selenium deposition amount of the target object when the planned feeding time is reached.

[0110] In some embodiments, the image processing module 402 includes:

[0111] A target detection unit is used to perform target detection on any target image to obtain at least one target detection frame, wherein the target image carries a device identification and acquisition time of the camera device to which it belongs, and the target detection frame is used to indicate any target object on the target image; an image segmentation unit is used to segment the target image into sub-image blocks corresponding to each target object based on the target detection frame; an image feature extraction unit is used to perform feature extraction on the sub-image blocks to obtain a feature map, and perform bilinear interpolation processing on the feature map to obtain a re-identification feature as the image feature of the target object; a spatiotemporal feature extraction unit is used to determine the spatiotemporal features of the target object indicated by the target detection frame based on the frame position data of the target detection frame on the target image and the acquisition time and device identification of the target image.

[0112] In some embodiments, the motion analysis module 403 includes:

[0113] A feature fusion unit is used to fuse the image features and spatiotemporal features of the target object contained in the target image to obtain the fusion features of the target image; a matching unit is used to traverse all target images for any target object, and match the detection results of the target object based on the fusion features, wherein the detection results are used to indicate the target images that capture the target object at multiple moments and the target detection frames that identify the target object on the target images; a target position determination unit is used to determine the target position data of the target object in the target image at each moment based on the frame position data of the target detection frames indicated by the detection results; a device position determination unit is used to determine the device position data of the corresponding camera device based on the device identification corresponding to the target image indicated by the detection results; a target tracking unit is used to determine the cumulative movement distance of the target object at each moment based on the target position data and the device position data, and generate the motion trajectory of the target object based on the cumulative movement distance.

[0114] In some embodiments, the matching unit includes:

[0115] An image division unit is used to divide the target image at a specified time into a first image, and divide the target images at other times except the specified time into a second image; an associated feature extraction unit is used to determine the first associated feature of the first image based on the fusion feature of the first image, and determine the second associated feature of the second image based on the fusion feature of the second image, wherein the first associated feature is used to indicate the degree of association between the target objects in the first image, and the second associated feature is used to indicate the degree of association between the target objects in any of the second images and the target objects in other second images; a similarity calculation unit is used to take any target object in the first image as a detection object, match the first associated feature for the detection object with the second associated feature corresponding to each target object in the second image, and determine the similarity between the detection object and the target object in the second image; a screening unit is used to screen out the target object in the second image corresponding to the similarity greater than a preset threshold as the matching object of the detection object; a detection result determination unit is used to determine the detection result of the detection object according to the target image to which the matching object belongs and the target detection frame on the target image used to indicate the matching object.

[0116] In some embodiments, the screening unit includes: a first matching object screening unit, which is used to use the screened target objects as the matching objects when the number of screened target objects does not exceed a preset number threshold; a second matching object screening unit, which is used to sort the similarities of the screened target objects from large to small when the number of screened target objects exceeds the number threshold, and select the first a target objects in the sort as the matching objects, where a is the number threshold.

[0117] In some embodiments, the associated feature extraction unit includes: a self-attention model processing unit, which is used to use a preset self-attention model to perform feature extraction on the fusion features of the first image to obtain a first associated feature; and a cross-attention model processing unit, which is used to use a preset cross-attention model to perform feature extraction on the fusion features of the second image to obtain a second associated feature.

[0118] In some embodiments, the environmental data includes temperature data, humidity data and lighting data of the pig house; the feeding data also includes feeding cycle data, feed formula data and feed additive data, the feeding cycle data includes the number of feedings per day and the amount of feeding each time, and the feed formula data includes the selenium content per unit mass in the feed.

[0119] In some embodiments, the prediction module 404 includes:

[0120] The first parameter acquisition unit is used to acquire a basic absorption rate, a weight coefficient, and feeding environment parameters related to the feeding process, wherein the feeding environment parameters include an environmental humidity parameter and an environmental temperature parameter, and the weight coefficient includes an exercise weight coefficient for characterizing the degree of influence of exercise on absorption efficiency, a temperature weight coefficient for characterizing the degree of influence of temperature on absorption efficiency, and a humidity weight coefficient for characterizing the degree of influence of humidity on absorption efficiency; an absorption rate model construction unit is used to construct an absorption rate model based on the basic absorption rate, the weight coefficient, the feeding environment data, the temperature data, the humidity data, and the exercise data determined by the motion trajectory, wherein the exercise data is used to characterize the cumulative moving distance of the target object; the second parameter acquisition unit is used to acquire a basic absorption rate, a weight coefficient, the feeding environment data, the temperature data, the humidity data, and the exercise data determined by the motion trajectory, and the exercise data is used to characterize the cumulative moving distance of the target object; An acquisition unit is used to obtain a basal metabolic consumption parameter, an exercise saturation distance, an additive effect parameter related to the feed additive data, and a consumption coefficient characterizing the increase in selenium consumption caused by oxidative stress; a selenium intake determination unit is used to determine the selenium intake based on the feeding cycle data, the feed formula data, and the feed additive data; a metabolic model construction unit is used to construct a metabolic model based on the basal metabolic consumption parameter, the exercise saturation distance, the additive effect parameter, the consumption coefficient, the selenium intake, and the absorption rate model; a training label generation unit is used to generate corresponding training labels according to the simulated deposition content output by the metabolic model; and a model training unit is used to train the selenium deposition prediction model using the training labels.

[0121] In some embodiments, the absorption rate model building unit includes:

[0122] The absorption rate model calculation unit is used to determine the absorption rate model by the following formula (1), including:

[0123] (1)

[0124] in, represents the absorptivity output by the absorptivity model, is the basic absorption rate, is the motion weight coefficient, is the exercise volume data, is the temperature weight coefficient, is the temperature data, is the ambient temperature parameter, is the humidity weight coefficient, is the humidity data, is the environmental humidity parameter;

[0125] In some embodiments, the metabolic model building unit comprises:

[0126] The metabolic model calculation unit is used to determine the metabolic model by the following formula (2), including:

[0127] (2)

[0128] in, represents the simulated deposition content output by the metabolic model, is the selenium intake, is the motion saturation distance, is the additive effect parameter, is the basal metabolic consumption parameter, is the consumption coefficient, is the temperature difference determined by the temperature data and the ambient temperature parameter, is a humidity difference determined by the humidity data and the ambient humidity parameter.

[0129] The device of the embodiments of the present application can execute the method provided by the embodiments of the present application, and the implementation principles are similar. The actions performed by each module in the device of each embodiment of the present application correspond to the steps in the method of each embodiment of the present application. For the detailed functional description of each module of the device, please refer to the description in the corresponding method shown in the previous text, which will not be repeated here.

[0130] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting selenium deposition in pigs, characterized in that: include: Acquire target images from multiple camera devices, environmental data of a pig house, and feeding data, wherein the target image is used to indicate at least one target object representing a pig, the feeding data includes planned feeding time, feeding cycle data, feed formula data, and feed additive data, and the environmental data includes temperature data, humidity data, and light data of the pig house; Performing image processing on the target image to obtain image features and spatiotemporal features of the target object contained in the target image, wherein the spatiotemporal features are used to indicate the position and appearance time of the target object in the target image; Determine the motion trajectory of each target object based on the image features and the spatiotemporal features; The environmental data of the target object at any time, the exercise data determined by the motion trajectory, and the feeding data are input into a pre-established selenium deposition prediction model to perform prediction, and the selenium deposition amount of the target object when the planned feeding time is reached is output, including: obtaining the basal absorption rate, the weight coefficient, and the feeding environment parameters related to the feeding process, wherein the feeding environment parameters include the environmental humidity parameter and the environmental temperature parameter, and the weight coefficient includes the exercise weight coefficient for characterizing the influence of the exercise amount on the absorption efficiency, the temperature weight coefficient for characterizing the influence of the temperature on the absorption efficiency, and the humidity weight coefficient for characterizing the influence of the humidity on the absorption efficiency; based on the basal absorption rate, the weight coefficient, the feeding environment parameters, the temperature data, the The humidity data and the exercise volume data determined by the motion trajectory are used to construct an absorption rate model, wherein the exercise volume data is used to characterize the cumulative moving distance of the target object; the basal metabolic consumption parameter, the exercise saturation distance, the additive effect parameter related to the feed additive data, and the consumption coefficient characterizing the increase in selenium consumption caused by oxidative stress are obtained; the selenium intake is determined based on the feeding cycle data, the feed formula data and the feed additive data; a metabolic model is constructed based on the basal metabolic consumption parameter, the exercise saturation distance, the additive effect parameter, the consumption coefficient, the selenium intake and the absorption rate model; according to the simulated deposition content output by the metabolic model, corresponding training labels are generated; and the selenium deposition prediction model is trained using the training labels.

2. The method for predicting selenium deposition in pigs according to claim 1, characterized in that: The performing image processing on the target image to obtain image features and temporal and spatial features of the target object contained in the target image includes: For any target image, target detection is performed on the target image to obtain at least one target detection frame, wherein the target image carries a device identification and acquisition time of a corresponding camera device, and the target detection frame is used to indicate any target object on the target image; According to the target detection frame, the target image is divided into sub-image blocks corresponding to each target object; Extracting features from the sub-image block to obtain a feature map, and performing bilinear interpolation processing on the feature map to obtain a re-identification feature as an image feature of the target object; Based on the frame position data of the target detection frame on the target image and the acquisition time and device identification of the target image, the spatiotemporal characteristics of the target object indicated by the target detection frame are determined.

3. The method for predicting selenium deposition in pigs according to claim 2, characterized in that: The determining the motion trajectory of each target object based on the image features and the spatiotemporal features includes: Fusing the image features and the temporal and spatial features of the target object contained in the target image to obtain fused features of the target image; For any target object, traverse all target images, and match the detection result of the target object based on the fusion feature, wherein the detection result is used to indicate the target image capturing the target object at multiple moments and the target detection frame identifying the target object on the target image; Determining target position data of the target object in the target image at each moment based on the frame position data of the target detection frame indicated by the detection result; Determine device location data of a corresponding camera device based on a device identifier corresponding to the target image indicated by the detection result; The cumulative moving distance of the target object at each moment is determined according to the target position data and the device position data, and the motion trajectory of the target object is generated according to the cumulative moving distance.

4. The method for predicting selenium deposition in pigs according to claim 3, characterized in that: For any target object, traversing all target images and matching the detection result of the target object based on the fusion feature includes: Classify the target image at a specified time as a first image, and classify the target image at other times except the specified time as a second image; Determine a first correlation feature of the first image based on the fusion feature of the first image, and determine a second correlation feature of the second image based on the fusion feature of the second image, wherein the first correlation feature is used to indicate the correlation between target objects in the first image, and the second correlation feature is used to indicate the correlation between target objects in any second image and target objects in other second images; Taking any target object in the first image as a detection object, matching a first correlation feature for the detection object with a second correlation feature corresponding to each target object in the second image, and determining a similarity between the detection object and the target object in the second image; Screening out the target object in the second image corresponding to the similarity greater than a preset threshold as a matching object for the detection object; A detection result of the detection object is determined according to a target image to which the matching object belongs and a target detection frame on the target image used to indicate the matching object.

5. The method for predicting selenium deposition in pigs according to claim 4, characterized in that: The step of selecting the target object in the second image corresponding to the target object having the similarity greater than a preset threshold as a matching object for the detection object includes: When the number of the screened target objects does not exceed the preset number threshold, the screened target objects are used as the matching objects; When the number of the screened target objects exceeds the quantity threshold, the similarities of the screened target objects are sorted from large to small, and the first a target objects in the sorting are selected as the matching objects, where a is the quantity threshold.

6. The method for predicting selenium deposition in pigs according to claim 5, characterized in that: The determining, based on the fusion feature of the first image, a first correlation feature of the first image, and determining, based on the fusion feature of the second image, a second correlation feature of the second image, includes: Using a preset self-attention model to extract features from the fusion features of the first image, to obtain first correlation features; The preset cross-attention model is used to extract the fusion features of the second image to obtain second associated features.

7. The method for predicting selenium deposition in pigs according to claim 6, characterized in that: The feeding cycle data include the number of feedings per day and the amount of feeding each time, and the feed formula data include the selenium content per unit mass in the feed.

8. The method for predicting selenium deposition in pigs according to claim 7, characterized in that: The step of inputting the environmental data of the target object at any time, the exercise data determined by the movement trajectory, and the feeding data into a pre-established selenium deposition prediction model to perform prediction and output the selenium deposition amount of the target object when the planned feeding time is reached comprises: The absorption rate model is determined by the following formula (1), including: E=E0·(1+k L ·Lk Te ·|You-You opt |-k H ·|HH opt |) (1) Wherein, E represents the absorption rate output by the absorption rate model, E0 is the basic absorption rate, k L is the motion weight coefficient, L is the motion amount data, k Te is the temperature weight coefficient, Te is the temperature data, Te opt is the ambient temperature parameter, k H is the humidity weight coefficient, H is the humidity data, and H opt is the environmental humidity parameter; The metabolic model is determined by the following formula (2), including: Among them, M net represents the simulated deposition content output by the metabolic model, I is the selenium intake, L max is the exercise saturation distance, θ is the additive effect parameter, R0 is the basal metabolic consumption parameter, k ROS is the consumption coefficient, ΔTe is the temperature difference determined by the temperature data and the ambient temperature parameter, and ΔH is the humidity difference determined by the humidity data and the ambient humidity parameter.

9. A device for predicting selenium deposition in pigs, characterized in that: include: A data acquisition module, used to acquire target images from multiple camera devices, environmental data of the pig house, and feeding data, wherein the target image is used to indicate at least one target object representing the pigs, the feeding data includes planned feeding time, feeding cycle data, feed formula data, and feed additive data, and the environmental data includes temperature data, humidity data, and light data of the pig house; An image processing module, used to perform image processing on the target image to obtain image features and spatiotemporal features of the target object contained in the target image, wherein the spatiotemporal features are used to indicate the position and appearance time of the target object in the target image; A motion analysis module, used to determine the motion trajectory of each target object based on the image features and the spatiotemporal features; The prediction module is used to input the environmental data of the target object at any time, the exercise data determined by the motion trajectory, and the feeding data into a pre-established selenium deposition prediction model to perform prediction and output the selenium deposition of the target object when the planned feeding time is reached, including: obtaining a basal absorption rate, a weight coefficient, and feeding environment parameters related to the feeding process, wherein the feeding environment parameters include an environmental humidity parameter and an environmental temperature parameter, and the weight coefficient includes an exercise weight coefficient for characterizing the degree of influence of the exercise amount on the absorption efficiency, a temperature weight coefficient for characterizing the degree of influence of the temperature on the absorption efficiency, and a humidity weight coefficient for characterizing the degree of influence of the humidity on the absorption efficiency; based on the basal absorption rate, the weight coefficient, the feeding environment parameters, the temperature parameters ... According to the feeding cycle data, the feed formula data and the feed additive data, an absorption rate model is constructed, and the exercise volume data is used to characterize the cumulative moving distance of the target object; the basal metabolic consumption parameter, the exercise saturation distance, the additive effect parameter related to the feed additive data and the consumption coefficient characterizing the increase in selenium consumption caused by oxidative stress are obtained; the selenium intake is determined based on the feeding cycle data, the feed formula data and the feed additive data; a metabolic model is constructed based on the basal metabolic consumption parameter, the exercise saturation distance, the additive effect parameter, the consumption coefficient, the selenium intake and the absorption rate model; according to the simulated deposition content output by the metabolic model, corresponding training labels are generated; and the selenium deposition prediction model is trained using the training labels.

Images