A feeding method and system based on holographic cognitive portrait

Abstract

The application discloses a kind of based on holographic cognitive portrait feeding method and system, the method includes collecting the facial image information of feed and recording the multi-dimensional biological characteristic data of feed, extract the facial image information feature of feed and calculate the identity of feed recognition;According to the multi-dimensional biological characteristic data of feed, construct the graph structure data of feed and calculate to obtain the feed label library, bind the feed label library and the identity of feed to generate the portrait of feed under dynamic update;Extract the data in the feed portrait and analyze the dynamic feeding requirement of feed, according to the feed ratio of the feeding of the feeding requirement adjustment.The feed ratio of the feeding of the application according to feeding requirement adjustment can realize accurate feeding.

Description

Technical Field

[0001] This invention relates to the field of animal husbandry technology, specifically to a feeding method and system based on holographic cognitive profiling. Background Technology

[0002] In recent years, internet companies and financial institutions have entered the smart farming industry. The introduction of digital technologies has led to a new look in the livestock sector, deeply integrating the internet with IoT (Internet of Things) and intelligent algorithms, driving the transformation from extensive to knowledge-based, technology-driven, and modern smart farming. Traditional methods for identifying dairy cows rely on manual reading or sensors. This requires writing or printing cow IDs on ear tags or placing sensors on the cow's surface, which can easily cause stress in the cows. Furthermore, manual reading methods are slow, and sensor-based identification methods are expensive.

[0003] Currently, facial recognition has been expanded from human face recognition to cow face recognition. The goal is to enable farmers to identify multiple dairy cows simultaneously and automatically without contact, reducing labor costs, avoiding stress reactions in the cows, and ensuring their health.

[0004] For example, application number 2021107383813 discloses a multi-target dairy cow identification method and system. The multi-target dairy cow identification method includes: determining a set of video frames to be identified based on a video to be identified; wherein the video to be identified includes facial image information of at least one dairy cow; obtaining and outputting a multi-target dairy cow facial image sequence based on the set of video frames to be identified and according to a multi-target detection and tracking network; and obtaining and outputting dairy cow identification information based on the multi-target dairy cow facial image sequence and according to a multi-target facial recognition network.

[0005] This patent application, while performing multi-target dairy cow identification, can only identify the cow's identity and cannot calculate the cow's current health status in real time. The specific health status still needs to be detected using traditional methods. Therefore, how to enable farmers to promptly detect health problems in cattle and remotely adjust the feed's nutrient composition to achieve precise feeding and improve production efficiency has become an urgent problem to be solved. Summary of the Invention

[0006] The purpose of this invention is to provide a feeding method and system based on holographic cognitive profiling to solve the problem in the prior art that feed cannot be adjusted according to the information of the animal.

[0007] To achieve the above objectives, the present invention is implemented using the following technical solution:

[0008] In a first aspect, the present invention discloses a feeding method based on holographic cognitive portraiture, comprising:

[0009] Collect facial images of animals and record multidimensional biological feature data of animals; extract facial image information features of animals and calculate and identify the identity of animals.

[0010] Based on the multidimensional biological characteristic data of the animals, we construct the animal graph structure data and calculate the animal tag library. We then bind the animal tag library and the animal identity to generate dynamically updated animal profiles.

[0011] Data is extracted from the animal profiles and the dynamic feeding needs of the animals are analyzed. The feed ratio is then adjusted according to these feeding needs.

[0012] Furthermore, collecting facial image information of the animals and recording their multidimensional biometric data, extracting facial image information features, and calculating and identifying the animals' identities include:

[0013] The YOLOv4 object detection algorithm is used to extract facial image information of animals from video frames;

[0014] The facial image information is input into the ResNet50 neural network for image feature extraction to obtain a feature map;

[0015] The feature map is decomposed into RGB three channels and then sparsely decomposed using the OMP algorithm. The database dictionary containing the animal information matrix is ​​updated based on the sparse decomposition results to identify the animal's identity.

[0016] Furthermore, based on the multidimensional biological characteristic data of the animals, a map structure data of the animals is constructed and a tag library of the animals is calculated. The animal tag groups and animal identities are bound to generate dynamically updated animal profiles, including:

[0017] Based on the multidimensional biological characteristic data of the animals and their corresponding time data, fill in and construct the animal graph structure data;

[0018] Tag clusters are designed based on different representations of the animal's condition to generate an animal tag library;

[0019] Based on the animal profile labeling results from low-backtracking spatial graph convolution, the animal tag library is bound to the animal identity to generate animal profiles with dynamically updated information.

[0020] Furthermore, extracting data from the animal profile and analyzing the animal's dynamic feeding needs, and adjusting the feed ratio according to these feeding needs, includes:

[0021] Extract animal tag clusters from the animal portrait, and extract and process the tag clusters to obtain the animal's dynamic feeding needs and optimal feeding goals;

[0022] Adjust the feed ratio, feeding time, and feeding amount according to the dynamic feeding needs and optimal feeding goals of the animals.

[0023] Furthermore, the multidimensional biometric data of the animals include: movement data, body temperature data, identity data, and physical condition data.

[0024] Secondly, this invention discloses a feeding system based on holographic cognitive profiling, comprising:

[0025] The body posture information recognition module is used to collect facial image information of animals and record multidimensional biological feature data of animals, extract facial image information features of animals and calculate and identify the identity of animals.

[0026] The profile building module is used to construct animal graph structure data based on multidimensional biological feature data of animals and calculate and obtain animal tag library. The animal tag library and animal identity are bound to generate dynamically updated animal profiles.

[0027] The feeding module is used to extract data from the animal profile and analyze the animal's dynamic feeding needs, and adjust the feed ratio according to the feeding needs.

[0028] Furthermore, the body posture information recognition module includes a data acquisition unit and an identity recognition unit; the portrait construction module includes a graph structure construction unit, a demand analysis unit, and a portrait generation unit; and the feeding module includes a demand analysis unit and a feeding terminal unit.

[0029] The data acquisition unit is used to extract animal facial image information from video frames using the YOLOv4 object detection algorithm, and input the facial image information into the ResNet50 neural network for image feature extraction to obtain a feature map; the identity recognition unit is used to perform RGB three-channel decomposition on the feature map, and perform sparse decomposition using the OMP algorithm, and update the database dictionary with animal information matrix according to the sparse decomposition result to identify the animal's identity.

[0030] The graph structure construction unit is used to fill and construct the animal graph structure data based on the multidimensional biological feature data of the animals and their corresponding time data; the requirement parsing unit is used to design tag clusters according to different representation forms of animal status to generate an animal tag library; the portrait generation unit is used to bind the animal tag library with the animal identity based on the animal portrait tag results of low backtracking spatial domain graph convolution to generate an animal portrait under dynamic information updates.

[0031] The demand analysis unit is used to extract animal tag clusters from the animal profile, extract and process the tag clusters to obtain the animal's dynamic feeding needs and optimal feeding goals; the feeding terminal unit is used to adjust the feed ratio, feeding timing and feeding amount according to the animal's dynamic feeding needs and optimal feeding goals.

[0032] Furthermore, the multidimensional biometric data of the animals include: movement data, body temperature data, identity data, and physical condition data.

[0033] Thirdly, the present invention discloses an electronic device, including a processor and a storage medium;

[0034] The storage medium is used to store instructions;

[0035] The processor is used to perform the steps of the above method according to the instructions.

[0036] Fourthly, the present invention discloses a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described above.

[0037] According to the above technical solution, the embodiments of the present invention have at least the following effects: The present invention constructs animal graph structure data based on multidimensional biological characteristic data of the animals and calculates and obtains an animal tag library, and binds the animal tag library and animal identity to generate dynamically updated animal portraits; it can accurately obtain the dynamic feeding needs of the animals, and adjust the feed ratio according to the feeding needs to achieve precise feeding. Attached Figure Description

[0038] Figure 1 This is an overall flowchart of the present invention;

[0039] Figure 2 This is a schematic diagram of the animal identification process in this invention;

[0040] Figure 3 A schematic diagram illustrating the process of creating animal portraits for this invention;

[0041] Figure 4 This is a schematic diagram of the feeding process in this invention. Detailed Implementation

[0042] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0043] This invention addresses the practical needs of current dairy farming, specifically targeting the crucial aspect of cattle feeding. It provides a key technology for intelligent precision feeding based on cattle facial recognition to solve the problems of incomplete sensory data, difficulty in analyzing cattle body condition, and inaccurate feeding decisions in current dairy farming. This includes: analyzing the dynamic feeding needs of cattle during their growth process based on individual information related to their behavioral characteristics and vital signs (such as weight, body temperature, and heart rate); developing a multi-functional guided feeding strategy based on intelligent feeding assistance; further designing a self-optimization mechanism to expand the effectiveness of precision feeding methods; and remotely adjusting the proportion of various nutrients in the cattle feed based on holographic images of the cattle uploaded by the system, thereby achieving remote management with separation of humans and livestock. Example 1

[0044] This invention discloses a feeding system based on holographic cognitive profiling, including a body posture recognition module, a profiling construction module, and a feeding module. The body posture recognition module is used to collect facial image information of the animal and record multidimensional biometric data of the animal, extract facial image information features of the animal, and calculate and identify the animal's identity. The profiling construction module is used to construct animal image structure data based on the multidimensional biometric data of the animal and calculate and obtain an animal tag library, and bind the animal tag library and animal identity to generate a dynamically updated animal profiling. The feeding module is used to extract data from the animal profiling and analyze the animal's dynamic feeding needs, and adjust the feed ratio according to the feeding needs.

[0045] The animals raised in this application can be livestock such as cattle and sheep, and this application will specifically use cattle as an example.

[0046] Specifically, the cattle body posture information recognition module of this application is used to collect facial image information of the animals and record multidimensional biometric data of the animals, extract facial image information features of the animals, and calculate and identify the identity of the animals. It can realize cross-scale feature fusion for cattle facial target detection and assess the overall health of cattle through parameter sharing.

[0047] The cattle profiling module is used to construct animal graph structure data based on multidimensional biological characteristic data of the animals and calculate and obtain an animal tag library. It then binds the animal tag library with the animal's identity to generate dynamically updated animal profiling. This allows for the fusion of multi-source biological information and mapping it to a high-dimensional state tag set for cattle, resulting in cattle profiling functions that facilitate clear feeding management decisions.

[0048] The feeding module extracts data from the animal profile and analyzes the animal's dynamic feeding needs, adjusting the feed ratio based on these needs. It enables remote adjustment of ambient temperature and humidity, as well as the proportion of various nutrients in the feed, and analyzes the dynamic feeding needs of cattle during their growth process, achieving precise feeding.

[0049] The specific research content for the above three modules is described below:

[0050] Cattle body posture information recognition module: This module uses the YOLOv4 target detection algorithm to extract facial image information of experimental cattle from video frames. The captured images are then input into a ResNet50 neural network for image feature extraction. Multidimensional biological characteristics and related parameters of the cattle are recorded to populate the cattle image structure data. This module includes a data acquisition unit, a target detection unit, and an identity recognition unit. The specific process is as follows: Figure 2 As shown.

[0051] Data Acquisition Unit: Data acquisition and feature extraction. First, the YOLOv4 object detection algorithm is used to extract facial image information of experimental cattle from video frames. At the same time, YOLOv4 is used to record the biological characteristics of cattle (retina, ear veins, facial feature information and related parameters). Next, in order to prevent overfitting, ResNet50 is used for secondary training. The captured images are input into the ResNet50 neural network for image feature extraction. This network introduces a three-layer residual module and outputs basic feature maps.

[0052] Comprehensive evaluation indicators: ;

[0053] in, Indicates the height of the feature map. Indicates the width of the feature map. This represents the number of channels in the feature map. Recall rate refers to the percentage of positive cases correctly predicted from the original sample.

[0054] Target Detection Unit: Multi-scale feature fusion target detection. To efficiently utilize feature maps of different scales, the YOLOv4 feature pyramid structure is used, enabling the network to automatically perform feature fusion across different feature layers. First, the previously acquired cattle movement data, body temperature data, identity data, body condition data, and their corresponding time data are used to construct a cattle map structure data. Requirement information is analyzed, semantic generalization is performed, and a high-precision cattle tag library is generated to achieve information fusion. A low-backtracking spatial domain map is used as the input to the target detection sub-network. Then, summation or max pooling operations are performed according to the actual situation of feature map resolution differences, and the results are retained in the maximized pooling output. Finally, the features recorded in the output pool are used to achieve accurate target localization of the cattle face in the overall captured image.

[0055] The identification unit, for cattle identification, performs RGB three-channel decomposition on the feature map containing the cattle face target, and then divides the generated image of each channel into a grid. The pixel values ​​of the image blocks are then unfolded into a column from left to right and top to bottom to obtain the reconstruction input matrix. Based on this, sparse decomposition is performed using the orthogonal matching pursuit algorithm, and the input matrix is ​​compared with the dictionary that previously stored the cattle information matrix. The atoms in the array are subjected to inner product operations one by one. An orthogonal projection is then performed on the direction of the atom with the largest inner product to construct a sparse approximation. Then, using... Second-rate The dictionary is updated by decomposition. Through two processes—cyclic sparse encoding and dictionary update—the process stops when the reconstruction error meets the accuracy requirements, and a corresponding training dictionary is generated in the dictionary database. , , These three training dictionaries will be used as the dictionaries for the next inner product operation. ;

[0056] The holographic perception-based cattle profiling module: By collecting multidimensional cattle data, it analyzes and integrates the previously acquired multi-source biological information, transforming the cattle difference map data into a unified alignment network structure to accommodate the variability of the cattle map data. This module mainly utilizes the previously completed graph structure construction unit and requirement analysis unit, and on this basis, implements the mapping unit and profiling generation unit, the specific relationships of which are as follows: Figure 3 As shown.

[0057] Graph Structure Construction Unit: Based on the previously acquired cattle movement data, body temperature data, identity data, body condition data, and their corresponding time data, the graph structure data of the cattle is constructed. Within the cattle graph structure dataset, for the graph set... Samples in and each corresponds to a K-dimensional vector representation of First, locate the prototype representation cluster of the K-dimensional cattle map data. A K-order approximation matrix is ​​calculated based on the distance between cattle point sets to align the node vector representation of each graph to the prototype representation cluster: ;

[0058] when and Distance between No. The smallest value in the row indicates the number of cattle in the chart data. Vector representation of the node and the first Alignment of prototype representations is achieved by aligning the representations of the cow graph data vertices with those of the common prototypes, thus obtaining transitive alignment information.

[0059] The requirements analysis unit analyzes and semantically generalizes specific requirements related to the cattle profiling in this project. For example, for the feeding management object "young and vigorous high-yielding, healthy, estrus-prone dairy cows," it generalizes terms such as "young and vigorous," "high-yielding," "healthy," and "estrus." It analyzes the semantic similarity and logical relevance between entities required for feeding management, calculates the distance between related terms, and obtains the relevance between terms and entities.

[0060] We designed a tagging system for feeding management, combining highly generalized strong semantic tags with highly accurate quantitative tags. We designed tag clusters according to different representations of cattle status, resulting in a well-organized and comprehensive cattle tag library.

[0061] Mapping Unit: After obtaining the transferable and consistent alignment information from the processed heterogeneous cattle data, cattle graph data of arbitrary size can be mapped to a fixed-size grid structure to fully utilize the irregular cattle data knowledge. Further calculation of the K-level alignment node feature matrix of the graph is then performed. and K-level aligned node adjacency matrix , constructed in Prototype diagrams showing pairwise similarity between K-dimensional prototype representations The K-dimensional prototype representation and matrix are reordered based on the similarity of edges and nodes between graphs.

[0062] A depth-based representation method is used to represent the cow node vectors to compute a K-level node consistency matrix, resulting in an aligned node mesh structure. Adjacency matrix of associated aligned mesh nodes By Transform into a low-backtracking adjacency matrix This is to avoid redundant information issues caused by undirected graph data of cattle.

[0063] Image generation unit: The network is divided into two parts. The first part represents a low-backtracking adjacency matrix of similarity. The vectors of two cow nodes are used as input to aggregate the vertex features of the node itself and its internal neighbors. Another set of cow node vectors is used as input to aggregate the vertex features of the node itself and its external neighbors. This yields a cow profile label result based on spatial graph convolution. The cow label group is then bound to the cow's identity, ultimately generating a dynamically updated cow profile. (Spatial graph convolution: performs a convolution operation on the node's neighborhood (weighted aggregation of information), and can also achieve parameter sharing.)

[0064] A precision feeding module based on holographic images of cattle: Based on the holographic images of cattle uploaded to the system, managers can remotely adjust the proportions of various nutrients in the feed according to the current condition of the cattle. This module includes a demand analysis unit and a feeding terminal unit, and its functions include... Figure 4 As shown.

[0065] Demand Analysis Unit: Based on the aforementioned holographic and precise cattle profiles, multi-layer features are extracted from the multi-dimensional tags of the cattle profiles, and the data is preprocessed to obtain a data sequence. This sequence is then used to analyze the dynamic feeding levels and optimal feeding targets of the cattle. Dynamic profile features and output data for different feeding routes are selected, such as feed intake, feeding frequency, feed ratio, milk production, and milk-to-feed ratio. The relevant features of the cattle feeding process are serialized according to time sequence, using one day or one week as the time step. Continuous and regular time-series feature data of cattle feeding processes over several days or weeks are selected for demand analysis. Based on the cattle profile tags, the dynamic feeding needs of the cattle during their growth process are analyzed, providing intelligent business support for the feeding process, thereby considering factors such as feed ratio, feeding timing selection, and feed quantity control.

[0066] Feeding terminal unit: Managers can remotely adjust the proportion of various nutrients in the feed according to the current condition of the cattle. While realizing remote management of humans and animals, this project is combined with Internet of Things technology to achieve precise feed delivery.

[0067] Demand Analysis Unit: This unit selects dynamic profile features and output data of cattle from different feeding routes, such as feed intake, feeding frequency, feed conversion ratio, milk production, and milk-to-feed ratio. The relevant features of the cattle feeding process are serialized according to time sequence, using one day or one week as the time step. Continuous and regular time-series feature data of cattle feeding processes over several days or weeks are selected for demand analysis. To this end, we establish two databases, DA1 and DA2, to store the acquired data. The feed conversion ratio adjustment process is as follows:

[0068] Step a: Establish the first database DA1. This database is used to store the farm identification dataset I, which can be used to uniquely identify a farm; it also stores the corresponding farm environmental data C. This forms corresponding data pairs.

[0069] Step b. Establish a second database DA2, which contains multiple standard documents [C, M, F, R, A, B], where C, M, F, R, A, and B represent environmental data, feed intake, feed frequency, feed ratio, milk yield, and milk-to-feed ratio, respectively. This database can be used to store the acquired serialized data. To determine the future relevant output data of a particular farm, first obtain the farm's identification data I. i And the expected milk production A i First, according to I i In DA1, the corresponding C is determined based on the data. i Then according to C i Determine the corresponding [M] k F k R k Bk The following error formula applies to C and A in DA2:

[0070] ;

[0071] in, and This represents the corresponding error value, which is a relatively small value. If there are more than one pair of data points for specific feed formulations, then the average value is taken. M k F k R k B k The expected feed intake, feeding frequency, feed formulation, and milk-to-feed ratio for this cycle.

[0072] Step c. Feed the cattle in the farm according to the algorithm determined above. During the feeding process, obtain the actual [M] of the feeding cycle. t F t A t B t ], and according to F t Determining R proportionally t Used to update DA2. M t F t R t A t B t This refers to the actual amount of feed consumed, feeding frequency, feed ratio, milk production, and milk-to-feed ratio for this cycle.

[0073] The algorithm described above is used to determine and update relevant feeding data in real time.

[0074] Based on cattle profile tags, the dynamic feeding needs of cattle during their growth process are analyzed to provide intelligent business support for the feeding process, thereby taking into account factors such as cattle feed ratio, feeding timing selection, and feed amount control.

[0075] Based on [M, F, R] in the second database DA2 mentioned above, namely feed intake, feeding frequency, and feed ratio, the feed ratio scheme and specific feeding time are determined. The feeding terminal unit is responsible for distributing instructions to the feed feeder of the corresponding cattle. By controlling the discharge time of different discharge outlets, the feed ratio is proportionally formulated. At the same time, feeding is carried out at corresponding intervals according to the feeding frequency. The feed intake in DA2 is the feeding amount, thereby achieving precise feeding.

[0076] This invention provides a multi-dimensional data acquisition model to improve the accuracy of cattle feeding. It promotes the development of the cattle breeding industry, continuously improves the level of precision in production management, saves breeding costs, and increases efficiency. It is beneficial to the transformation from traditional breeding models to intelligent livestock farming, accelerates the modernization of the dairy industry, and can effectively improve the precision and efficiency of feeding management while reducing labor costs. Example 2

[0077] Based on the same inventive concept as Embodiment 1, this embodiment discloses a feeding method based on holographic cognitive profiling, including: Step 1, collecting facial image information of the animal and recording multidimensional biometric data of the animal, extracting facial image information features of the animal and calculating and identifying the animal's identity. Step 2, constructing animal image structure data based on the multidimensional biometric data of the animal and calculating and obtaining an animal tag library, binding the animal tag library and animal identity to generate a dynamically updated animal profile. Step 3, extracting data from the animal profile and analyzing the animal's dynamic feeding needs, adjusting the feed ratio according to the feeding needs.

[0078] This invention constructs animal graph structure data based on multidimensional biological characteristic data of animals and calculates and obtains an animal tag library. The animal tag library and animal identity are bound to generate dynamically updated animal portraits. It can accurately obtain the dynamic feeding needs of animals and adjust the feed ratio according to the feeding needs to achieve precise feeding.

[0079] Step 1: Data collection and feature extraction, feature fusion and target detection, and cattle identification.

[0080] Step 11: First, use the YOLOv4 object detection algorithm to extract the facial image information of the experimental cattle from the video frames, and use the YOLOv4 object detection algorithm to record the multidimensional biological characteristics of the cattle (cattle movement data, body temperature data, identity data, body condition data) and their related parameters.

[0081] The YOLOv4 target detection algorithm used in this application is a single-stage target detection algorithm. Based on YOLOv3, this algorithm adds some new improvement ideas, which greatly improves its speed and accuracy.

[0082] Step 1.2: Input the captured facial image into the ResNet50 neural network for image feature extraction.

[0083] The ResNet50 neural network is a structure that can effectively solve the degradation problem of deep networks. It transforms the fitting of multiple nonlinear changes in the input into fitting the residual between the input and the output, thus becoming an identity mapping. 50 means 50 layers.

[0084] Step 1.3: Perform RGB three-channel decomposition on the detected cow face targets. Based on this, use OMP (Orthogonal Matching Pursuit Algorithm) for sparse decomposition. Identify the cow's identity through two processes: cyclic sparse coding and database dictionary update.

[0085] OMP stands for Orthogonal Matching Pursuit. This application uses this algorithm, which is simple and efficient.

[0086] Step 2: Fill in the data to construct the cattle image structure, perform semantic generalization on specific requirement information, and generate cattle images under dynamically updated information.

[0087] Step 2.1: Based on the previously obtained cattle movement data, body temperature data, identity data, body condition data and their corresponding time data, populate and construct the cattle graph structure data.

[0088] Step 2.2: Analyze the specific requirements information and perform semantic generalization. Analyze the semantic similarity and logical correlation between entities required for feeding management, and sort out the complex cattle information to generate a high-precision cattle tag library.

[0089] Step 2.3: Utilize the two parts of the graph network structure to aggregate the vertex features of the node itself and its internal neighbors, and then aggregate the vertex features of the node itself and its external neighbors. This yields the cattle portrait labeling results based on low-backtracking spatial graph convolution. The cattle label group is then bound to the cattle identity, ultimately generating a dynamically updated cattle portrait.

[0090] Step 3: Extract data from the images, analyze the dynamic feeding needs of the cattle, and adjust the feed ratio for precise feeding.

[0091] Step 3.1: Based on the cattle portrait obtained above, extract multi-layer features from the multi-dimensional labels of the cattle portrait and preprocess the data to obtain a data sequence.

[0092] Step 3.2: Based on the cattle profile tags, analyze the dynamic feeding needs of cattle during their growth process, and adjust the feed ratio, feeding timing, and feeding amount accordingly.

[0093] Step 3.3: Remotely adjust the proportion of various nutrients in the feed according to the current condition of the cattle to achieve precise feeding.

[0094] This invention provides a multi-dimensional data acquisition model to improve the accuracy of cattle feeding. It promotes the development of the cattle breeding industry, continuously improves the level of precision in production management, saves breeding costs, and increases efficiency. It is beneficial to the transformation from traditional breeding models to intelligent livestock farming, accelerates the modernization of the dairy industry, and can effectively improve the precision and efficiency of feeding management while reducing labor costs. Example 3

[0095] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0096] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0097] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0098] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0099] The embodiments of this application also provide an electronic device, which may be a tablet computer, a smartphone, a personal digital assistant, etc.

[0100] Electronic devices may include: memory, processor, communication interface and communication bus, the communication bus being used to enable communication between these components.

[0101] The memory is used to store all model data, as well as various data such as the calculation program instructions corresponding to the feeding method and system based on the whole cognitive profile provided in the embodiments of this application. The memory can be random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable read-only memory (EPROM), etc.

[0102] When the processor reads and runs computer program instructions stored in memory that correspond to the feeding method based on the holistic cognitive profile, it executes the emergency decision control method provided in the embodiments of this application.

[0103] A processor may be an integrated circuit chip with signal processing capabilities. The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), discrete gate or transistor logic devices, or discrete hardware components.

[0104] As is known from common technical knowledge, this invention can be implemented through other embodiments that do not depart from its spirit or essential characteristics. Therefore, the disclosed embodiments described above are merely illustrative in all respects and are not the only ones. All modifications within the scope of this invention or its equivalents are included in this invention.

Claims

1. A feeding method based on holographic cognitive profiling, characterized in that, include: Collect facial images of animals and record multidimensional biological feature data of animals; extract facial image information features of animals and calculate and identify the identity of animals. Based on the multidimensional biological characteristic data of the animals, we construct the animal graph structure data and calculate the animal tag library. We then bind the animal tag library and the animal identity to generate dynamically updated animal profiles. Extract data from the animal profile and analyze the dynamic feeding needs of the animals, and adjust the feed ratio according to the feeding needs; Based on the multidimensional biological characteristic data of the animals, a map structure data of the animals is constructed and a tag library of the animals is calculated. The animal tag groups are bound to the animal identities to generate dynamically updated animal profiles, including: Based on the multidimensional biological characteristic data of the animals and their corresponding time data, fill in and construct the animal graph structure data; Tag clusters are designed based on different representations of the animal's condition to generate an animal tag library; Based on the animal profile labeling results of low backtracking spatial graph convolution, the animal tag library is bound to the animal identity to generate animal profiles with dynamic information updates. Extracting data from the animal profiles and analyzing the animals' dynamic feeding needs, and adjusting the feed ratio based on these feeding needs, includes: Extract animal tag clusters from the animal portrait, and extract and process the tag clusters to obtain the animal's dynamic feeding needs and optimal feeding goals; Adjust the feed ratio, feeding time, and feeding amount according to the dynamic feeding needs and optimal feeding goals of the animals.

2. The feeding method based on holographic cognitive profiling according to claim 1, characterized in that, Collecting facial images of captive animals and recording their multidimensional biometric data, extracting facial image features, and calculating and identifying the animals' identities include: The YOLOv4 object detection algorithm is used to extract facial image information of animals from video frames; The facial image information is input into the ResNet50 neural network for image feature extraction to obtain a feature map; The feature map is decomposed into RGB three channels and then sparsely decomposed using the OMP algorithm. The database dictionary containing the animal information matrix is ​​updated based on the sparse decomposition results to identify the animal's identity.

3. The feeding method based on holographic cognitive profiling according to claim 1, characterized in that, The multidimensional biometric data of the animals include: movement data, body temperature data, identity data, and physical condition data.

4. A feeding system based on holographic cognitive profiling, characterized in that, include: The body posture information recognition module is used to collect facial image information of animals and record multidimensional biological feature data of animals, extract facial image information features of animals and calculate and identify the identity of animals. The profile building module is used to construct animal graph structure data based on multidimensional biological feature data of animals and calculate and obtain animal tag library. The animal tag library and animal identity are bound to generate dynamically updated animal profiles. The feeding module is used to extract data from the animal profile and analyze the dynamic feeding needs of the animal, and adjust the feed ratio according to the feeding needs; the profile construction module includes a graph structure construction unit, a demand parsing unit and a profile generation unit, and the feeding module includes a demand analysis unit and a feeding terminal unit; the graph structure construction unit is used to fill and construct animal graph structure data based on the multidimensional biological characteristic data of the animal and its corresponding time data; The requirement parsing unit is used to design tag clusters based on different representations of the animal's state to generate an animal tag library; The image generation unit is used to bind the animal tag library with the animal identity based on the animal image tag results of low backtracking spatial graph convolution, and generate animal images with dynamic information updates. The demand analysis unit is used to extract animal tag clusters from the animal profile, extract and process the tag clusters to obtain the animal's dynamic feeding needs and optimal feeding goals; the feeding terminal unit is used to adjust the feed ratio, feeding timing and feeding amount according to the animal's dynamic feeding needs and optimal feeding goals.

5. The feeding system based on holographic cognitive profiling according to claim 4, characterized in that, The body posture information recognition module includes a data acquisition unit and an identity recognition unit; The data acquisition unit is used to extract animal facial image information from video frames using the YOLOv4 object detection algorithm, and input the facial image information into the ResNet50 neural network for image feature extraction to obtain a feature map; the identity recognition unit is used to perform RGB three-channel decomposition on the feature map, and perform sparse decomposition using the OMP algorithm, and update the database dictionary with animal information matrix according to the sparse decomposition result to identify the animal's identity.

6. The feeding system based on holographic cognitive profiling according to claim 4, characterized in that, The multidimensional biometric data of the animals include: movement data, body temperature data, identity data, and physical condition data.

7. An electronic device, characterized in that, Including processor and storage media; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1 to 3.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method according to any one of claims 1 to 3.

Images