Individualized feeding method and device based on single cage rearing

Abstract

The embodiment of the invention aims to provide a personalized feeding method and device based on single cage rearing, a computer program product and a computer program storage medium. The method comprises the following steps: acquiring multi-modal characteristic data of a single cage rearing object, including static characteristics and dynamic characteristics; and inputting the multi-modal feature data into a multi-modal deep learning model and predicting a corresponding feeding scheme. According to the invention, individualized precise feeding of cage-cultured poultry is realized, so that the breeding management is changed from extensive type to refined type, and a differentiated management means is provided for a high-end farm. The feed waste is effectively reduced through accurate feeding, the health problem caused by excessive or insufficient nutrition supply is avoided, and the growth speed and production benefits of the poultry are improved. In addition, real-time updating of the dynamic optimization strategy template ensures that the feeding strategy is always matched with the actual demand, so that the economic benefit is maximized.

Description

Technical Field

[0001] The various embodiments of the present disclosure relate to the fields of artificial intelligence and aquaculture technology, and in particular to a technology for personalized feeding based on single-cage farming. Background Art

[0002] In caged poultry farming, such as large-scale broiler or laying hen operations in integrated cages, traditional feeding methods typically employ standardized batch management strategies. These strategies standardize feed amounts and formulations based on average individual needs, utilizing centralized feeding equipment for automated operations. However, this approach lacks the ability to precisely tailor feed to individual differences. This can lead to over- or undernutrition in some individuals, particularly when different individuals within a flock exhibit varying nutritional requirements, further impacting growth performance, health, and production efficiency.

[0003] In order to solve this problem, some existing technologies attempt to collect basic individual data, such as weight changes or feeding behavior, by installing weight sensors, cameras and other equipment in chicken cages. However, most of these data are used for basic grouping or abnormality monitoring, and lack in-depth analysis of individual feeding needs. Existing technologies usually rely on rule sets or empirical models for simple parameter adjustments, such as dividing different feeding amount intervals according to weight ranges. However, this static adjustment strategy is difficult to respond to changes in dynamic environmental factors (such as temperature and humidity) and individual behavioral characteristics (such as feed intake and activity level) in real time, resulting in limited optimization capabilities of feeding plans. Summary of the Invention

[0004] The purpose of each embodiment of the present disclosure is to provide a personalized feeding method, device, computer program product and computer program storage medium based on single-cage breeding.

[0005] According to one aspect of the present disclosure, a personalized feeding method based on single cage culture is provided, wherein the method comprises the following steps:

[0006] Collecting multimodal feature data of a single caged subject, including static features and dynamic features; wherein the static features include the age, weight, and body surface image of the single caged subject; and the dynamic features include data related to the single caged subject's food intake, activity level, and physiological state, as well as real-time environmental data;

[0007] The multimodal feature data is input into a multimodal deep learning model and a corresponding feeding plan is predicted; wherein,

[0008] generating an initial feeding plan for the single caged object according to the static characteristics;

[0009] Generating a dynamic optimization strategy template for the single caged object according to the initial feeding plan and the dynamic characteristics;

[0010] A final feeding plan is generated according to the initial feeding plan and the dynamic optimization strategy template.

[0011] According to one aspect of the present disclosure, a personalized feeding device based on single-cage breeding is also provided, wherein the device includes a memory and a processor, the memory storing computer program instructions, and when the computer program instructions are executed by the processor, the device is configured to perform the following operations:

[0012] Collecting multimodal feature data of a single caged subject, including static features and dynamic features; wherein the static features include the age, weight, and body surface image of the single caged subject; and the dynamic features include data related to the single caged subject's food intake, activity level, and physiological state, as well as real-time environmental data;

[0013] The multimodal feature data is input into a multimodal deep learning model and a corresponding feeding plan is predicted; wherein,

[0014] generating an initial feeding plan for the single caged object according to the static characteristics;

[0015] Generating a dynamic optimization strategy template for the single caged object according to the initial feeding plan and the dynamic characteristics;

[0016] A final feeding plan is generated according to the initial feeding plan and the dynamic optimization strategy template.

[0017] According to one aspect of the present disclosure, a computer program product is further provided, comprising computer program instructions, wherein when the computer program instructions are executed by a computer device, the computer device is configured to perform the following operations:

[0018] Collecting multimodal feature data of a single caged subject, including static features and dynamic features; wherein the static features include the age, weight, and body surface image of the single caged subject; and the dynamic features include data related to the single caged subject's food intake, activity level, and physiological state, as well as real-time environmental data;

[0019] The multimodal feature data is input into a multimodal deep learning model and a corresponding feeding plan is predicted; wherein,

[0020] generating an initial feeding plan for the single caged object according to the static characteristics;

[0021] Generating a dynamic optimization strategy template for the single caged object according to the initial feeding plan and the dynamic characteristics;

[0022] A final feeding plan is generated according to the initial feeding plan and the dynamic optimization strategy template.

[0023] According to one aspect of the present disclosure, a computer program storage medium is further provided, wherein computer executable instructions are stored. When the computer executable instructions are executed by a computer device, the computer device is configured to perform the following operations:

[0024] Collecting multimodal feature data of a single caged subject, including static features and dynamic features; wherein the static features include the age, weight, and body surface image of the single caged subject; and the dynamic features include data related to the single caged subject's food intake, activity level, and physiological state, as well as real-time environmental data;

[0025] The multimodal feature data is input into a multimodal deep learning model and a corresponding feeding plan is predicted; wherein,

[0026] generating an initial feeding plan for the single caged object according to the static characteristics;

[0027] Generating a dynamic optimization strategy template for the single caged object according to the initial feeding plan and the dynamic characteristics;

[0028] A final feeding plan is generated according to the initial feeding plan and the dynamic optimization strategy template.

[0029] The embodiments of the present disclosure combine multimodal deep learning technology with modern farming equipment to propose a personalized feeding solution based on the fusion of static and dynamic features, which not only solves the shortcomings of traditional feeding methods but also has a profound impact on the modern farming industry.

[0030] This disclosure demonstrates unique innovation in terms of technical implementation:

[0031] Deep fusion of static and dynamic features: A CNN extracts deep features from poultry surface images and combines them with static features such as weight and age to generate personalized initial feeding plans, enabling fundamental analysis of individual differences. Transformer-based modeling of the complex nonlinear relationship between dynamic features (such as feed intake, activity level, and environmental data) and feeding requirements generates a dynamic optimization strategy template, including feeding adjustment rules, feature weight matrices, and response curve parameters. This deep fusion of multimodal data breaks the limitations of traditional analysis that relies on a single feature.

[0032] Application of self-supervised learning: Using self-supervised learning to train the Transformer avoids the tedious sample labeling process, while achieving efficient mining of dynamic feature patterns, enabling dynamic optimization strategy templates to adapt to changes in the environment and individual behavior in real time.

[0033] Refined control of decision-making optimization: A multi-layer perceptron (MLP) is used to fuse the initial feeding plan with the dynamic optimization strategy template, outputting the final plan including feeding amount, feed formula, and ratio. This multi-stage optimization strategy further improves feeding accuracy.

[0034] This disclosure also brings the following positive impacts to the aquaculture industry:

[0035] Improvement of segmented management: The traditional group management model has been broken through. This disclosure realizes individualized and precise feeding of caged poultry, which transforms the breeding management from "extensive" to "fine", providing differentiated management methods for high-end farms.

[0036] Optimizing breeding returns: Precision feeding effectively reduces feed waste, avoids health problems caused by excessive or insufficient nutrition, and improves poultry growth and production efficiency. Furthermore, real-time updates to dynamic optimization strategy templates ensure that feeding strategies always match actual needs, maximizing economic benefits.

[0037] Data-driven intelligent upgrade: By collecting and analyzing dynamic characteristics, this disclosure provides data support for poultry farmers, helping them understand the growth status of individual poultry and groups, and make data-driven farming decisions, thereby promoting the development of modern farming industry towards intelligent and sustainable directions.

[0038] Environmental friendliness and optimized resource utilization: Precision feeding reduces excessive feed consumption and excrement production, helping to lower the risk of environmental pollution on farms and providing support for green farming. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Other features, objects and advantages of the present disclosure will become more apparent from a reading of the detailed description of non-limiting embodiments made with reference to the following drawings:

[0040] Figure 1 A flow chart showing a personalized feeding method based on single-cage breeding according to an exemplary embodiment of the present disclosure is shown.

[0041] The same or similar reference numerals in the drawings represent the same or similar components. DETAILED DESCRIPTION

[0042] The specific embodiments of the present disclosure will be further described below with reference to the accompanying drawings.

[0043] Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments of the present disclosure are described as devices represented by block diagrams and processes or methods represented by flow charts. Although the flow charts describe the operating processes of the various embodiments of the present disclosure as sequential processing, many of the operations therein can be implemented in parallel, concurrently or simultaneously. In addition, the order of the operations can be rearranged. The processes of the various embodiments of the present disclosure can be terminated when their operations are completed, but can also include additional steps not shown in the flow charts. The processes of the various embodiments of the present disclosure can correspond to methods, functions, procedures, subroutines, subprograms, etc.

[0044] The methods illustrated by the flowcharts and the devices illustrated by the block diagrams discussed below may be implemented by hardware, software, firmware, middleware, microcode, hardware description languages, or any combination thereof. When implemented in software, firmware, middleware, or microcode, the program code or code segments that perform the necessary tasks may be stored in a machine or computer-readable medium such as a storage medium. (One or more) processors may perform the necessary tasks.

[0045] Similarly, it will also be understood that any flow charts, flow diagrams, state transition diagrams, and the like represent various processes that can be fully described as program code stored in a computer-readable medium and thereby executed by a computer device or processor, whether or not such computer device or processor is explicitly shown.

[0046] As used herein, the term "storage medium" may refer to one or more devices for storing data, including read-only memory (ROM), random access memory (RAM), magnetic RAM, kernel memory, magnetic disk storage media, optical storage media, flash memory devices, and / or other machine-readable media for storing information. The term "computer-readable medium" may include, but is not limited to, portable or fixed storage devices, optical storage devices, and various other media capable of storing and / or containing instructions and / or data.

[0047] A code segment can represent a procedure, function, subroutine, program, routine, subroutine, module, software package, class, or any combination of instructions, data structures, or program descriptions. A code segment can be coupled to another code segment or hardware circuit by passing and / or receiving information, data, arguments, parameters, or memory contents. Information, arguments, parameters, data, etc. can be passed, forwarded, or transmitted via any suitable means, including shared memory, message passing, token passing, network transmission, etc.

[0048] In this context, "computer device" refers to an electronic device that can perform predetermined processing procedures such as numerical calculations and / or logical calculations by running predetermined programs or instructions. It may include at least a processor and a memory, wherein the processor executes program instructions pre-stored in the memory to perform the predetermined processing procedure, or the predetermined processing procedure is performed by hardware such as ASIC, FPGA, DSP, or a combination of the above two.

[0049] The above-mentioned "computer device" is usually expressed in the form of a general-purpose computer device, and its components may include but are not limited to: one or more processors or processing units, system memory. The system memory may include a computer-readable medium in the form of a volatile memory, such as a random access memory (RAM) and / or a cache memory. The "computer device" may further include other removable / non-removable, volatile / non-volatile computer-readable storage media. The memory may include at least one computer program product having a set (e.g., at least one) program modules that are configured to perform the functions and / or methods of the various embodiments of the present disclosure. The processor executes various functional applications and data processing by running the programs stored in the memory.

[0050] For example, a computer program for executing various functions and processes of the various embodiments of the present disclosure is stored in the memory. When the processor executes the corresponding computer program, the various embodiments of the present disclosure are implemented.

[0051] Typically, a computer device may be, for example, a user device or a network device, or even a combination of the two. The user device includes, but is not limited to, a personal computer (PC), a laptop computer, a mobile terminal, etc., and the mobile terminal includes, but is not limited to, a smartphone, a tablet computer, etc.; the network device includes, but is not limited to, a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing (Cloud Computing) consisting of a large number of computers or network servers, wherein cloud computing is a type of distributed computing, a super virtual computer consisting of a group of loosely coupled computer sets. The computer device may be run independently to implement the various embodiments of the present disclosure, or may be connected to a network and implement the various embodiments of the present disclosure through interactive operations with other computer devices in the network. The network in which the computer device is located includes, but is not limited to, the Internet, a wide area network, a metropolitan area network, a local area network, a VPN network, etc.

[0052] It should be noted that the user devices, network devices and networks are merely examples. Other existing or future computing devices or networks that are applicable to the embodiments of the present disclosure should also be included in the scope of protection of the present disclosure and are incorporated herein by reference.

[0053] The specific structural and functional details disclosed herein are merely representative and are for the purpose of describing exemplary embodiments of the present disclosure. However, the various embodiments of the present disclosure may be implemented in many alternative forms and should not be construed as being limited to only the embodiments set forth herein.

[0054] It should be understood that although the terms "first," "second," and the like may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used solely to distinguish one element from another. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element, without departing from the scope of the exemplary embodiments. The term "and / or" as used herein includes any and all combinations of one or more of the listed associated items.

[0055] The terms used herein are intended only to describe specific embodiments and are not intended to limit exemplary embodiments. Unless the context clearly indicates otherwise, the singular forms "a", "an", "an item" used herein are also intended to include the plural. It should also be understood that the terms "comprise" and / or "include" used herein specify the presence of stated features, integers, steps, operations, units and / or components, and do not preclude the presence or addition of one or more other features, integers, steps, operations, units, components and / or combinations thereof.

[0056] It should also be noted that, in some alternative implementations, the functions / actions mentioned may occur in a different order than that indicated in the accompanying drawings. For example, two figures shown in succession may actually be performed substantially simultaneously or may sometimes be performed in the reverse order, depending on the functions / actions involved.

[0057] See Figure 1 , which shows a flow chart of a personalized feeding method based on single cage breeding according to an embodiment of the present disclosure.

[0058] like Figure 1 As shown, in step S1, the computer device collects multimodal feature data of a single caged object, including static features and dynamic features; in step S2, the computer device inputs the multimodal feature data into a multimodal deep learning model and predicts a corresponding feeding plan.

[0059] Individual cage farming is a modern farming management model in which individual poultry or livestock are housed in separate farming units, typically individual cages or isolated spaces. Each cage or unit provides dedicated space and resources for each animal, ensuring personalized feeding. This farming method offers significant advantages in improving feed utilization, reducing disease transmission, and facilitating daily management. It provides precise individual management and monitoring, ensuring a standardized and precise farming environment.

[0060] Individual animals or poultry, typically chickens, rabbits, and other farmed animals, are housed in a single cage system. In this system, each animal is treated as a separate management unit, with personalized adjustments made to its growth, feeding, health monitoring, and feeding. The management approach for individual animals emphasizes precise tracking and adjustment of each individual animal, including tailored feeding plans and management measures based on its growth status, health status, and food intake.

[0061] In a caged environment, individual caged animals are usually separated and raised independently. Identifying each individual is the key to achieving personalized feeding, ensuring that multimodal feature data can be accurately matched to a specific individual, while supporting personalized management and long-term data tracking.

[0062] To identify individual caged subjects, each can be implanted with an RFID tag or attached to a cage, with its unique identification information read via a reader / writer. Alternatively, a QR code or barcode can be attached to an ear tag, foot ring, or cage body, and scanned with a camera to obtain the subject's identity information. Furthermore, cage numbers can be associated with individual caged subjects, allowing data on individual caged subjects to be managed in a fixed, linked manner.

[0063] For ease of explanation, the description of personalized feeding plans in this disclosure often uses single chickens as individual caged objects for example and specific description. The feeding plans provided in this disclosure can be feeding amounts, feed formulas and ratios, or feeding amounts and feed types.

[0064] The present disclosure does not restrict the structure of individual cage systems. For example, when using integrated chicken cages for large-scale farming, each integrated chicken cage includes multiple compartments (individual cages), each with its own independent feed trough. Based on the personalized feeding plan disclosed herein, a personalized feeding plan is generated for each chicken, and feeding is completed in its corresponding feed trough.

[0065] Here, a feasible way is to realize personalized feeding of each partition unit by setting up a partition silo and an independent feeding pipe. The silo includes a plurality of partition silos, each of which stores a feed raw material (such as corn, soybean) or a feed type (such as complete pellet feed, grass meal feed). The feed in the silo can be transported to the corresponding chicken cage through an independent feeding pipe according to the personalized feeding plan generated by the control system in accordance with the present disclosure. Another feasible way is to consider having a robot be responsible for feeding, that is, the robot takes food from the aforementioned silo according to the personalized feeding plan and delivers it to the corresponding chicken cage. The robot feeding plan is more suitable for scenarios where the partition units are dispersed.

[0066] Specifically, in step S1, a computer device collects multimodal feature data of a single caged object, including static features and dynamic features.

[0067] Here, the static features include the age, weight, and body surface image of the individual caged objects.

[0068] Static characteristics refer to relatively fixed or slowly changing characteristics, which are used to describe the basic properties of individual caged objects. These data are mainly obtained through regular collection and recording, and include the following:

[0069] 1) Age: Record the birth date or number of days the individual caged chickens have lived as an important basis for assessing their growth stage and formulating feeding plans. For example, different growth stages of broilers require different feed formulas, and the age of breeders affects their reproductive performance.

[0070] 2) Weight: This data is measured regularly using a weighing device and used to assess individual growth and health. This data can help determine feed conversion efficiency, optimize feed intake, and promptly identify abnormal weight deviations from the normal range.

[0071] 3) Body surface images: A camera is used to capture high-definition images of the subject's body surface. Computer vision techniques (such as CNN) are then used to extract image features, such as feather sheen, skin color, and body contours. These features can reflect the subject's health, such as nutritional deficiencies or illnesses.

[0072] Dynamic features include data related to feed intake, activity level, and physiological status of individual caged objects, as well as real-time environmental data.

[0073] Dynamic characteristics refer to the real-time changing characteristics of individual caged animals, covering multiple dimensions such as behavior, feeding, and physiology, and can reflect their immediate status and needs. These data are usually collected by sensors, cameras, or other real-time monitoring equipment, and include the following:

[0074] 1) Feed intake data

[0075] Eating frequency: The frequency of an individual's eating is monitored through a feeding trough sensor or camera to reflect their appetite and eating habits.

[0076] Feeding time: The duration of each feeding session, reflecting the chicken's demand for feed.

[0077] Feed intake: The actual amount of food consumed by an animal over a certain period of time is recorded using a weighing or other sensor device. This, combined with the type of feed, allows for an assessment of whether nutrient intake is adequate or wasteful.

[0078] 2) Activity data

[0079] Movement intensity: Sensors or cameras analyze the birds' movement patterns to determine their activity intensity, such as how active they are or how long they're still. Abnormal activity levels may indicate health issues, such as illness or injury.

[0080] Activity time: records the total activity time of the chicken to help determine the metabolic level of the chicken.

[0081] 3) Physiological status related data

[0082] Body temperature: collected through non-contact infrared sensors or contact sensors to monitor whether there is fever or hypothermia.

[0083] 4) Real-time environmental data

[0084] Temperature and humidity: Temperature and humidity data in the cage and surrounding environment are used to determine whether they meet the optimal breeding conditions.

[0085] Light intensity: Recorded by light sensors, light time and intensity are optimized according to the poultry's circadian rhythm.

[0086] Air quality: Gas sensors monitor ammonia, carbon dioxide concentrations, etc. to ensure good air conditions and reduce the risk of respiratory diseases.

[0087] In step S2, the computer device inputs the multimodal feature data into a multimodal deep learning model and predicts a corresponding feeding plan.

[0088] The feeding scheme provided in the present disclosure may be:

[0089] 1) Feeding amount, feed formula and ratio

[0090] For example: Feeding amount: 110 g / day; Feed formula and proportion: (corn: 50%; soybean: 30%; fish meal: 15%; premix and additives: 5%)

[0091] 2) Feeding amount and feed type

[0092] For example: feeding amount: 120 g / day; feed type: complete pellet feed (designed specifically for laying hens, containing a high calcium formula).

[0093] By training the multimodal deep learning model, the present disclosure can provide the above two feeding schemes for individual caged objects. The feeding scheme 1) is specifically described below in this article.

[0094] Here, the multimodal deep learning model further generates a personalized feeding plan for the current single caged object through the following steps.

[0095] Step 1): Generate an initial feeding plan based on the static characteristics of the individual caged objects.

[0096] The numerical features of age and weight are normalized and mapped to a uniform range for easier model processing. Surface image features are extracted using a CNN to obtain a deep representation related to health status and body shape. The extracted image features are concatenated with the normalized numerical features and fed into a fully connected layer for fusion calculations, ultimately generating an initial feeding plan, including feed amount, formula, and ratio.

[0097] Step 2): Generate a dynamic optimization strategy template based on the initial feeding plan and the dynamic characteristics of the individual caged objects.

[0098] Here, the transformer structure can be used to generate a dynamic optimization strategy template based on the initial feeding plan and the dynamic characteristics of the individual caged objects.

[0099] The collected dynamic features (feed intake, activity level, environmental data, etc.) are input into the Transformer model along with the initial feeding plan. The Transformer model effectively processes time series data, captures relationships between features, and optimizes and adjusts future dynamic trends related to feeding.

[0100] Dynamic optimization policy templates include:

[0101] -Rule sets, including feeding adjustment rules based on each dynamic feature.

[0102] The rule set defines feeding adjustment rules based on dynamic characteristics such as feed intake, activity level, and environmental data. These rules help make appropriate adjustments based on changes in different dynamic characteristics. For example, low feed intake may require an increase in feed amount, while high ambient temperature may require an adjustment to the feed formula.

[0103] - Weight matrix, used to represent the contribution weight of each dynamic feature to feeding adjustment.

[0104] Different dynamic characteristics have different degrees of influence on feeding requirements. The weight matrix helps to determine which characteristics have a greater impact on adjusting the requirements, so as to formulate a more accurate feeding plan.

[0105] -Response curve parameters, describing the relationship between dynamic characteristics and feeding requirements.

[0106] Response curve parameters describe the relationship between these dynamic characteristics and feeding requirements. These dynamic characteristics include feed intake, activity level, environmental data, and more. Response curve parameters can reflect how these factors collectively influence changes in feeding requirements. This relationship helps optimize feeding strategies to better meet the actual needs of individual birds, thereby improving feeding accuracy and efficiency. Therefore, response curves can quantify the connection between dynamic characteristics and feeding requirements, predict how changes in dynamic characteristics will affect future feeding requirements, and thus provide a theoretical basis for subsequent feeding optimization.

[0107] The Transformer model consists of an encoder and a decoder. The specific structure is as follows:

[0108] Encoder part:

[0109] Input: initial feeding plan and dynamic characteristics of individual caged animals (such as feed intake, activity level, physiological status, environmental data, etc.).

[0110] Function: The encoder accepts the concatenation of the initial feeding scheme and dynamic features, captures the relationship between different time steps and features through the self-attention mechanism, and generates feature representations. This part mainly extracts the timing and context information from the input data.

[0111] Processing: In each encoder layer, a high-level semantic representation of the input data is gradually constructed through a multi-head self-attention mechanism and a feed-forward neural network.

[0112] Decoder part:

[0113] Input: The input of the decoder includes the feature representation generated by the encoder and the target generation information of the previous step. The encoder output is combined with the current state of the decoder through the cross-attention mechanism.

[0114] Function: The decoder generates a dynamic optimization strategy template (including rule sets, weight matrices, response curve parameters, etc.) based on the dynamic features and current state information extracted by the encoder. Each generation step depends on the previous generation results to gradually adjust the feeding plan.

[0115] Processing: Similar to the encoder, the decoder also processes input features through a self-attention mechanism and fuses information with previously generated content. The error between the generated output and the expected target is used as the loss during training to guide the optimization of the Transformer model.

[0116] Function: Through the encoder-decoder structure, the Transformer model can generate a corresponding dynamic optimization strategy template after processing the input initial feeding plan and dynamic features, ensuring that subsequent feeding plans can be accurately adjusted and optimized.

[0117] This structure can well capture the complex interactions between temporal correlations and multimodal features in the input data, thereby optimizing personalized feeding strategies.

[0118] Here, the transformer structure can be trained through self-supervised learning, and the training goal of the transformer structure is to minimize the error between the dynamic optimization strategy template and the expected target.

[0119] Self-supervised learning is an unsupervised learning method in which the Transformer model learns the structure and patterns in the data through generative tasks. Specifically, based on the input initial feeding plan and dynamic characteristics, the Transformer model outputs a dynamically optimized strategy template that meets the aquaculture needs. This strategy template consists of a set of rules, a weight matrix, and response curve parameters. The goal of the Transformer model is to minimize the error between the generated strategy template and the actual needs.

[0120] Training task definition:

[0121] Input: The input of the Transformer model includes the initial feeding plan (such as the feeding amount, feed formula and ratio for the day) and dynamic characteristics (such as feed intake, activity level, environmental data, etc.).

[0122] Target output: The Transformer model needs to output a "dynamic optimization strategy template," which is a set of data containing feeding adjustment rules, weight matrices, and response curve parameters based on dynamic features.

[0123] In each training step, the Transformer model inputs the initial feeding plan and dynamic features to generate a dynamic optimization strategy template.

[0124] Calculate the loss function and optimize the model through gradient descent. The loss function calculates the error between the generated dynamic optimization strategy template (rule set, weight matrix, response curve) and the actual target. Specifically:

[0125] Rule set error: A classification loss function is used to measure the difference between the predicted rules and the true rules.

[0126] Weight Matrix Error: The difference between the predicted weight matrix and the actual weight matrix can be measured using the mean squared error (MSE).

[0127] Response curve error: The mean square error can also be used to measure the difference between the predicted response curve and the actual demand.

[0128] Optimize the Transformer model parameters through backpropagation and optimization algorithms (such as Adam or SGD) to minimize the loss function.

[0129] The training is repeated until the Transformer model converges and the generated dynamic optimization strategy template can match the expected results as accurately as possible.

[0130] Self-supervised learning, as a training method, enables the Transformer model to generate dynamic optimization strategy templates by inputting initial feeding plans and dynamic features without explicit labels. By optimizing the error between the generated strategy templates and actual requirements, the Transformer model's performance is gradually improved. Ultimately, the trained Transformer model is able to automatically generate highly adaptable and effective dynamic optimization strategy templates based on input data (static and dynamic features) from actual farming scenarios.

[0131] Step 3): Generate the final feeding plan based on the initial feeding plan and the dynamic optimization strategy template.

[0132] Here, a multi-layer perceptron (MLP) can be used to generate a final feeding plan based on the initial feeding plan and the dynamic optimization strategy template.

[0133] The MLP structure is as follows:

[0134] Input layer: All features of the initial feeding plan and the dynamic optimization strategy template are concatenated to form a complete input vector. This vector may contain multiple numerical features and embedded representations (such as rule sets, weight matrices, response curve parameters, etc.).

[0135] Hidden layers: A series of fully connected layers perform nonlinear transformations to extract complex relationships in the data. Each layer uses an activation function (such as ReLU) to introduce nonlinear transformations, helping the model learn more complex feature representations.

[0136] Output layer: Outputs the final feeding plan, including the daily feeding amount, feed formula, and ratio. This layer usually has multiple output nodes, each corresponding to a specific feeding parameter (e.g., feeding amount, corn ratio, soybean ratio, etc.).

[0137] The final output feeding plan is a precisely adjusted feeding amount, feed formula, and ratio. This plan can be dynamically adjusted based on the initial feeding plan and dynamic optimization strategy template to optimize feeding results and production efficiency.

[0138] For the multimodal deep learning model as a whole, the loss function of the deep learning model needs to consider the impact of static and dynamic features on the final feeding plan, as well as the quality of the output at each step.

[0139] The loss function of a multimodal deep learning model can be expressed as a weighted combination of multiple parts:

[0140] L = α L 初始 + β L transformer + γ L 最终 (1)

[0141] Among them, α, β, and γ are weight coefficients used to balance the contribution of the losses of different tasks to the total loss. The loss values ​​of the same task may be of different magnitudes. Appropriate adjustment of the weights helps optimize the training effect.

[0142] L 初始 is the loss of the initial feeding plan, as follows:

[0143]

[0144] in, represents the initial feeding plan generated by the model;

[0145] Q 目标 Indicates the target feeding plan;

[0146] MSE: Mean squared error, used to measure the prediction error of feeding amount, feed formula and ratio;

[0147] L transformer To dynamically optimize the loss of the strategy template, the details are as follows:

[0148] L transformer = λ1 L 规则集 + λ2 L 权重矩阵 + λ3 L 响应曲线 (3)

[0149] Among them, λ1, λ2, and λ3 are weight coefficients used to balance the contribution of each part and need to be selected according to the Transformer model tuning;

[0150] L 规则集 For the rule set loss, you can use categorical cross entropy or KL divergence to evaluate whether the rules match;

[0151] L 权重矩阵 For the weight matrix loss, MSE can be used to evaluate the prediction quality of each dynamic feature weight;

[0152] L 响应曲线 For the response curve loss, the difference or mean square error of the curve fit can be used to measure the deviation between the predicted and actual response curves;

[0153] L 最终 The loss of the final feeding plan is as follows:

[0154]

[0155] in, Represents the final feeding plan generated by the model;

[0156] Q 目标 Indicates the target feeding plan;

[0157] MSE: Mean square error, used to minimize the error between the final feeding plan and the actual demand.

[0158] Therefore, a feasible training process for multimodal deep learning models can include two stages:

[0159] Separate training phase: Separate the model parts of each task to obtain the initial weights and task capabilities.

[0160] Joint optimization stage: Joint training is performed based on the loss function L, and α, β and γ are adjusted simultaneously to coordinate multi-task performance.

[0161] Through this design, the entire multimodal deep learning model can effectively learn from static and dynamic features, gradually optimize the feeding plan, and meet actual needs.

[0162] It should be noted that the various embodiments of the present disclosure may be implemented in software and / or a combination of software and hardware, for example, may be implemented using an application specific integrated circuit (ASIC), a general purpose computer, or any other similar hardware device. In one embodiment, the software programs of the various embodiments of the present disclosure may be executed by a processor to implement the steps or functions described above. Similarly, the software programs of the various embodiments of the present disclosure (including related data structures) may be stored in a computer-readable recording medium, for example, a RAM memory, a magnetic or optical drive, or a floppy disk and the like. In addition, some steps or functions of the various embodiments of the present disclosure may be implemented using hardware, for example, as a circuit that cooperates with a processor to perform various steps or functions.

[0163] In addition, at least a portion of the various embodiments of the present disclosure may be implemented as a computer program product, such as computer program instructions, which, when executed by a computing device, can invoke or provide the methods and / or technical solutions according to the various embodiments of the present disclosure through the operation of the computing device. The program instructions for invoking / providing the methods of the various embodiments of the present disclosure may be stored in a fixed or removable recording medium, and / or transmitted via a data stream in a broadcast or other signal-carrying medium, and / or stored in the working memory of a computing device that operates according to the program instructions.

[0164] For those skilled in the art, it is obvious that the embodiments of the present disclosure are not limited to the details of the above-mentioned exemplary embodiments, and the embodiments of the present disclosure can be implemented in other specific forms without departing from the spirit or basic characteristics of the embodiments of the present disclosure. Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-restrictive, and the scope of the embodiments of the present disclosure is limited by the appended claims rather than the above description, and it is intended that all changes that fall within the meaning and scope of the equivalent elements of the claims are included in the embodiments of the present disclosure. Any figure mark in the claims should not be regarded as limiting the claims involved. In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the system claim can also be implemented by one unit or device through software or hardware. Words such as first and second are used to indicate names and do not indicate any specific order.

Claims

1. A personalized feeding method based on single cage culture, wherein: The method comprises the following steps: Collecting multimodal feature data of a single caged subject, including static features and dynamic features; wherein the static features include the age, weight, and body surface image of the single caged subject; and the dynamic features include data related to the single caged subject's food intake, activity level, and physiological state, as well as real-time environmental data; The multimodal feature data is input into a multimodal deep learning model and a corresponding feeding plan is predicted; wherein, generating an initial feeding plan for the single caged object according to the static characteristics; Generating a dynamic optimization strategy template for the single caged object according to the initial feeding plan and the dynamic characteristics; A final feeding plan is generated according to the initial feeding plan and the dynamic optimization strategy template.

2. The method according to claim 1, wherein The dynamic optimization strategy template includes: - a set of rules, including feeding adjustment rules based on each of the dynamic characteristics; - a weight matrix, used to represent the contribution weight of each of the dynamic characteristics to the feeding adjustment; - Response curve parameters, describing the relationship between the dynamic characteristics and feeding requirements.

3. The method according to claim 1 or 2, wherein: A transformer structure trained by self-supervised learning is used to generate the dynamic optimization strategy template according to the initial feeding plan and the dynamic characteristics; the training goal of the transformer structure of the self-supervised learning is to minimize the error between the dynamic optimization strategy template and the expected target. A Transformer structure that has undergone self-supervised learning is used to generate a dynamic optimization strategy template based on the initial feeding plan and dynamic features. The goal of the self-supervised learning task is to minimize the error between the dynamic optimization strategy template and the expected target.

4. The method according to claim 1, wherein A multilayer perceptron is used to generate a final feeding plan based on the initial feeding plan and the dynamic optimization strategy template.

5. The method according to claim 1, wherein The feeding plan includes feeding amount, feed formula and ratio.

6. The method according to claim 1, wherein The feeding plan includes feeding amount and feed type.

7. The method according to claim 1, wherein The single caged object is caged poultry.

8. A personalized feeding device based on single cage culture, wherein: The device includes a memory and a processor, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the device is configured to perform the following operations: Collecting multimodal feature data of a single caged subject, including static features and dynamic features; wherein the static features include the age, weight, and body surface image of the single caged subject; and the dynamic features include data related to the single caged subject's food intake, activity level, and physiological state, as well as real-time environmental data; The multimodal feature data is input into a multimodal deep learning model and a corresponding feeding plan is predicted; wherein, generating an initial feeding plan for the single caged object according to the static characteristics; Generating a dynamic optimization strategy template for the single caged object according to the initial feeding plan and the dynamic characteristics; A final feeding plan is generated according to the initial feeding plan and the dynamic optimization strategy template.

9. A computer program product comprising computer program instructions, wherein: When the computer program instructions are executed by a computer device, the computer device is configured to perform the method according to any one of claims 1 to 7.

10. A computer program storage medium having computer executable instructions stored therein, wherein when the computer executable instructions are executed by a computer device, the computer device is configured to perform the method according to any one of claims 1 to 7.

Images