A method and system for optimizing ecological breeding of white-feathered broiler chickens

Abstract

The present invention relates to the technical field of breeding management, and discloses a method and system for optimizing ecological breeding of white-feathered broilers, comprising the following steps: dividing a flock of white-feathered broilers into a healthy group, a slightly lame group, and a severely lame group, collecting record information of each group and multidimensional record information of the flock every hour; generating plane data No. 1 based on the record information collected at the three most recent consecutive time points, and generating plane data No. 2 based on the collected multidimensional record information of the flock; inputting the plane data No. 1 and the plane data No. 2 into a feeding strategy model, and outputting the feeding strategies representing each group; sampling the crop fullness rate of the flock every hour, and when the crop fullness rate of the sampled chickens is lower than 95%, regenerating the feeding strategies of each group; and adjusting the grouping of the flock every other day. The beneficial effect of the present invention is that the present invention can alleviate leg problems and reduce mortality while maintaining a high crop fullness rate by grouping white-feathered broilers.

Description

Technical Field

[0001] The present invention relates to the technical field of breeding management, and more specifically, to a method and system for optimizing the ecological breeding of white-feathered broiler chickens. Background Art

[0002] In modern farming, white-feathered broilers are widely farmed because of their rapid growth and excellent meat quality. However, farming efficiency is closely related to the health of the chickens. When raising white-feathered broilers, the feeding method is very important, and the feeding method can be dynamically changed through the crop fullness rate. The crop is an important part of the chicken's digestive system, responsible for storing and initially digesting feed. After the chicken eats, the food will first enter the crop for storage, and then gradually enter the subsequent digestive organs. Due to different factors such as the age of the chicken and the environment, the crop fullness rate will also be different. The fullness of the crop directly affects the chicken's nutrient absorption, growth rate and overall health. Therefore, it is necessary to maintain a high crop fullness rate by changing the feed feeding plan.

[0003] Because white-feathered broiler chickens grow rapidly, their weight will increase rapidly in a short period of time, which puts pressure on their bones and joints. In the breeding environment, the chickens have little space to move around and lack of exercise, which will also affect the healthy growth of their muscles and bones.

[0004] Due to the aggravation of leg problems, chickens may suffer from some leg diseases in the subsequent breeding process, and untimely treatment of leg problems may lead to increased mortality of chickens. Therefore, it is necessary to consider the leg health of chickens in the breeding process and adopt a more reasonable feeding strategy. Summary of the invention

[0005] The present invention provides a method and system for optimizing the ecological breeding of white-feathered broilers, which solves the technical problem in the related art that white-feathered broilers grow too fast, resulting in greater pressure on leg muscles and bones, leading to subsequent leg diseases and low survival rate.

[0006] The present invention provides a method for optimizing ecological breeding of white-feathered broiler chickens, comprising the following steps:

[0007] Step 100, the white-feathered broiler chickens are artificially divided into three groups: a healthy group, a slightly lame group, and a severely lame group, and the record information of each group and the multi-dimensional record information of the chicken group are collected every hour, the record information includes feeding records, chicken group information, and environmental information; the multi-dimensional record information of the chicken group includes: chicken group information, environmental information, and chicken group image information;

[0008] In a preferred embodiment, the feeding record includes: protein content, fat content, calcium content, phosphorus content, vitamin content, feeding amount and feeding frequency of each group feed;

[0009] Chicken flock information includes: the number, average weight and age of each group of chickens;

[0010] Environmental information includes: temperature, humidity, and light;

[0011] The chicken flock image information includes: high-resolution images of each group of chicken flocks.

[0012] Step 200, generating plane data No. 1 based on the record information collected at the three most recent consecutive time points, and generating plane data No. 2 based on the collected multi-dimensional record information of the chicken flock;

[0013] In a preferred embodiment, when the collected data is less than three times, artificial experience is used for feeding.

[0014] In a preferred embodiment, the first plane data includes a first bin for storing data, a serial data stream of a first bin indexing a first data object, and the first plane data also includes a first relationship list for storing association relationships between the first data objects indexed by the first bin, and the element value of the i-th row and j-th column of the first relationship list represents the quantized value of the association relationship between the first data objects indexed by the i-th and j-th first bins;

[0015] The association relationship includes high and low, corresponding to two quantitative values ​​respectively.

[0016] Data objects No. 1 include flocks, flock groups, environment, and feeding records;

[0017] The first data objects with high association relationships are defined as: chicken flocks and their environment, chicken flocks and the flock group they belong to, and feeding records and the flocks they belong to;

[0018] A serial data stream of a data object No. 1 includes n data units sorted by time, and the tth data unit represents the record information collected at the tth moment of the data object.

[0019] In a preferred embodiment, the second plane data includes a second facet for storing data, one second facet indexes one second data object, and the second plane data also includes a second relationship list for storing association relationships between the second data objects indexed by the second facet, and the element value of the i-th row and j-th column of the second relationship list represents the quantitative value of the association relationship between the second data objects indexed by the i-th and j-th second facets;

[0020] Data objects No. 2 include chicken flocks, chicken flock groups, environments, and chicken flock images;

[0021] The second data object with a high correlation relationship is defined as: a flock of chickens and its environment, a flock of chickens and the flock group to which it belongs, and a flock of chickens in a flock image including the flock of chickens in the flock image.

[0022] Step 300, inputting the first plane data and the second plane data into a feeding strategy model, and outputting a feeding strategy representing each group;

[0023] In a preferred embodiment, the feeding strategy model includes a first encoder, a second encoder, a third encoder, a first feature fusion layer, a first fully connected layer and a second fully connected layer, wherein the first encoder is used to input plane data No. 1 and output a first identification feature to the second encoder, the second encoder outputs a second identification feature to the first feature fusion layer, the third encoder inputs plane data No. 2 and outputs a third identification feature to the first feature fusion layer and the second fully connected layer, the second fully connected layer outputs a leg health status of each group, the first feature fusion layer outputs a first fusion feature to the first fully connected layer, and the first fully connected layer outputs a feeding strategy of each group.

[0024] In a preferred embodiment, the multi-dimensional recorded information of the chicken flock is input into the model after feature engineering, and the multi-dimensional recorded information of the chicken flock includes image modality, and the CNN algorithm is used as a feature engineering method for the image modality information.

[0025] Step 400, sampling the crop fullness rate of the chickens every hour, and when the crop fullness rate of the sampled chickens is lower than 95%, regenerating the feeding strategy for each group;

[0026] Step 500, adjust the grouping of the chicken flock every other day.

[0027] In a preferred embodiment, the third encoder is connected to the second fully connected layer for independent pre-training, and the training calculation formula is as follows:

[0028] ;

[0029] in represents the loss value, represents the total number of training samples, represents the true leg health score annotated by the expert for the qth sample, Represents the leg health score of the qth sample output by the model.

[0030] In a preferred embodiment, the training of the feeding strategy model is performed using reinforcement learning, and the training steps of the feeding strategy model are as follows:

[0031] Step 101, copy the feeding strategy model to generate an Actor network and a TargetActor network, use a feedforward neural network as a Critic network, copy the Critic network to obtain a TargetCritic network, and randomly initialize the parameters of the Actor network, the TargetActor network, the Critic network, and the TargetCritic network;

[0032] Step 102, input The recorded information collected Go to the Actor network to get the Feeding strategy implemented ;

[0033] No. Feeding strategy implemented Add random noise to get Noise-enhanced feeding strategy ;

[0034] The first Feeding strategy implemented Input into the environment and get immediate rewards from feedback and The recorded information collected ; Experience , , , Deposit into experience pool;

[0035] Step 103, randomly obtain N experiences from the experience pool;

[0036] enter Go to the TargetActor network to get the Feeding strategy implemented ;

[0037] Step 104, calculate the first loss value and the second loss value , gradient descent method updates the Actor network and Critic network;

[0038] ;

[0039] ;

[0040] ;

[0041] in Represents input and The output of the Critic network is Indicates execution The comprehensive Q value when express and The output of the TargetCritic network;

[0042] ;

[0043] in, express The crop fullness rate of the chickens at all times, represents the target crop fullness rate, express The number of healthy chickens at all times, represents the total number of chickens, express The cost of the strategy adopted at any time, represents the total cost, Indicates immediate reward;

[0044] is the discount factor, which is a value between 0 and 1, used to balance the weights of immediate rewards and future rewards. The default value is 0.6;

[0045] pass Update the Actor network by Update the Critic network;

[0046] Step 105, every fixed number of network updates, update the TargetActor so that its parameters are the same as those of the current Actor network, and update the TargetCritic network so that its parameters are the same as those of the current Critic network;

[0047] Step 106, until the Actor network and the Critic network converge, the step is terminated, or after the number of cycles of steps 103-105 reaches a set value, the Actor network replaces the feeding strategy model and the step is terminated.

[0048] The beneficial effect of the present invention is that the present invention groups white-feathered broilers according to leg problems and manages feeding strategies for each group respectively. The model takes into account the leg health problems of the flock. Through different feeding strategies for each group, not only a high crop fullness rate can be maintained from the feed aspect, but also leg problems can be effectively alleviated, thereby reducing leg disease problems and mortality in the flock. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 The present invention is a flow chart of an ecological breeding optimization method for white-feathered broiler chickens. DETAILED DESCRIPTION

[0050] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that the discussion of these implementations is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the present specification. Various examples may omit, replace, or add various processes or components as needed. In addition, the features described in some examples may also be combined in other examples.

[0051] At least one embodiment of the present invention discloses a method for optimizing ecological breeding of white-feathered broiler chickens, such as Figure 1 As shown, the following steps are included:

[0052] Step 100, the white-feathered broiler chickens are artificially divided into three groups: a healthy group, a slightly lame group, and a severely lame group, and the record information of each group and the multi-dimensional record information of the chicken group are collected every hour, the record information includes feeding records, chicken group information, and environmental information; the multi-dimensional record information of the chicken group includes: chicken group information, environmental information, and chicken group image information;

[0053] Feeding records include: protein content, fat content, calcium content, phosphorus content, vitamin content, feed amount, and feeding frequency in each group feed;

[0054] Chicken flock information includes: the number, average weight and age of each group of chickens;

[0055] Environmental information includes: temperature, humidity, and light;

[0056] The chicken flock image information includes: high-resolution images of each group of chicken flocks;

[0057] Step 200, generating plane data No. 1 based on the record information collected at the three most recent consecutive time points, and generating plane data No. 2 based on the collected multi-dimensional record information of the chicken flock;

[0058] The first plane data includes a first bin for storing data, a serial data stream of a first bin indexing a first data object, and a first relationship list for storing association relationships between the first data objects indexed by the first bin, wherein the element value of the i-th row and j-th column of the first relationship list represents the quantitative value of the association relationship between the first data objects indexed by the i-th and j-th first bins;

[0059] The association relationship includes high and low, corresponding to two quantitative values ​​respectively.

[0060] Data objects No. 1 include flocks, flock groups, environment, and feeding records;

[0061] The first data objects with high association relationships are defined as: chicken flocks and their environment, chicken flocks and the flock group they belong to, and feeding records and the flocks they belong to;

[0062] A serial data stream of a data object No. 1 includes n data units sorted by time, and the tth data unit represents the record information collected at the tth moment of the data object.

[0063] The second plane data includes a second facet for storing data, one second facet indexes a second data object, and the second plane data also includes a second relationship list for storing association relationships between the second data objects indexed by the second facet, and the element value of the i-th row and j-th column of the second relationship list represents the quantitative value of the association relationship between the second data objects indexed by the i-th and j-th second facets;

[0064] Data objects No. 2 include chicken flocks, chicken flock groups, environments, and chicken flock images;

[0065] The second data object with high association relationship is defined as: chickens and their environment, chickens and the chicken group to which they belong, and chickens in the image of chickens including chickens in the image of chickens;

[0066] In one embodiment of the present invention, when the collected data is less than three times, artificial experience is used for feeding.

[0067] In one embodiment of the present invention, the feeding record can be represented as a vector {{22, 6, 0.87, 0.7, 12000, 110, 2}, {21, 5, 1.2, 0.8, 15000, 100, 2}, {20, 4, 1.5, 1, 20000, 60, 3}}, the three vector components represent three groups, the groups represent the healthy group, the mild lameness group and the severe lameness group, and the values ​​in the vector represent the protein content: 22%, the fat content: 6%, the calcium content: 0.87%, the phosphorus content: 0.7%, and the vitamin C content: 1. Vegetarian content: 12,000 IU per kilogram, feed feeding amount: 110 grams per chicken, feeding frequency: twice a day; Protein content: 21%, fat content: 5%, calcium content: 1.2%, phosphorus content: 0.8%, vitamin content: 15,000 IU per kilogram, feed feeding amount: 100 grams per chicken, feeding frequency: twice a day; Protein content: 20%, fat content: 4%, calcium content: 1.5%, phosphorus content: 1%, vitamin content: 20,000 IU per kilogram, feed feeding amount: 90 grams per chicken, feeding frequency: three times a day.

[0068] Step 300, inputting the No. 1 plane data and the No. 2 plane data into a feeding strategy model, the feeding strategy model comprising a first encoder, a second encoder, a third encoder, a first feature fusion layer, a first fully connected layer, and a second fully connected layer, wherein the first encoder is used to input the No. 1 plane data, output a first identification feature to the second encoder, the second encoder outputs a second identification feature to the first feature fusion layer, the third encoder inputs the No. 2 plane data, outputs a third identification feature to the first feature fusion layer and the second fully connected layer, the second fully connected layer outputs a leg health status representing each group, the first feature fusion layer outputs a first fusion feature to the first fully connected layer, and the first fully connected layer outputs a feeding strategy representing each group;

[0069] In one embodiment of the present invention, the calculation formula of the first encoder is as follows:

[0070] ;

[0071] ;

[0072] ;

[0073] ;

[0074] ;

[0075] ;

[0076] in represents the tth data unit of the serial data stream corresponding to the vth first bin of the first plane data, and They respectively represent the t-th and t-1-th first identification features of the serial data stream corresponding to the v-th bin number 1 of the plane data number 1, , , , and They represent the tth first, second, third, fourth and fifth intermediate features of the vth first face element of the first plane data, respectively. represents the t-1th fourth intermediate feature of the vth first face element of the first plane data, → represents the circulation direction, , , , , , , , Respectively represent the first, second, third, fourth, fifth, sixth, seventh, and eighth serial recognition weight parameters, , , , Respectively represent the first, second, third, and fourth serial identification bias parameters represents the Hadamard product, , n represents the total number of data units in the serial data stream, when t=1 , tanh is the tanh function, Represents the Sigmoid function.

[0077] In one embodiment of the present invention, the calculation formula of the second encoder is as follows:

[0078] ;

[0079] in represents the second identification feature of the vth first facet in the lth layer, represents the set of No. 1 facets with high correlation with No. 1 facet v, represents the first plane recognition weight matrix of layer l, is the attention weight of the first facet u on the first facet v at the lth layer, which is calculated by the following formula:

[0080] ;

[0081] in is a learnable attention vector, represents the vector concatenation operation, represents the second plane recognition weight matrix of layer l, represents transpose, yes Activation function;

[0082] and They represent the facet features of the vth facet number one in the l-1th layer, , E represents the total number of layers, when l=1 , , when l=E is equal to the identification feature of the first bin v; and They respectively represent the t-th and t-1-th first identification features of the serial data stream corresponding to the v-th and u-th first bins of the first plane data, and n represents the total number of data units of the serial data stream.

[0083] In one embodiment of the present invention, the calculation formula of the third encoder is as follows:

[0084] ;

[0085] ;

[0086] in represents the aggregate representation of the vth second facet at layer l, and denote the third identification features of the vth second facet at the lth layer and the l-1th layer respectively, represents the set of second facets with high correlation with the second facet v; Represents the universal quantifier symbol; represents the facet feature of the u-th facet No. 2 at the l-1th layer, , E represents the total number of layers, when l=1 , , , Respectively represent the feature representation of the second data object indexed by the vth second facet and the uth second facet, when l=E is equal to the third identification feature of the second face element v, Represents the aggregation function of layer l, such as average or maximum pooling function. represents the plane recognition weight matrix of the lth layer, σ represents the Sigmoid function, Represents the concatenation function.

[0087] In one embodiment of the present invention, the calculation formula of the second fully connected layer is as follows:

[0088] ;

[0089] in represents a vector, the i-th component of the vector represents the leg health score of the i-th group of chickens, represents the third identification feature of the vth second facet of the second plane data, N represents the set of the second facets of all the second plane data, Represents the second full connection.

[0090] In one embodiment of the present invention, the calculation formula of the first feature fusion layer is as follows:

[0091] ;

[0092] in, represents the first fusion feature, represents the concatenation function, represents the sum weight matrix, represents the summation bias parameter, represents the second identification feature of the vth first facet in the lth layer, represents the set of No. 1 facets with high correlation with No. 1 facet v, represents the third identification feature of the vth second facet in the lth layer, It represents the set of second facets that have a high correlation with the second facet v.

[0093] In one embodiment of the present invention, the calculation formula of the first fully connected layer is as follows:

[0094] ;

[0095] in Represents a vector, the i-th component of the vector represents the feeding strategy of the i-th group of chickens, a feeding strategy is represented by a vector, the components of the vector represent the protein content, fat content, calcium content, phosphorus content, vitamin content, feed amount, feeding frequency, represents the first fusion feature, Represents the first full connection.

[0096] In one embodiment of the present invention, multi-dimensional recorded information of a flock of chickens is input into a model after feature engineering, and the multi-dimensional recorded information of a flock of chickens includes image modalities, and the information of the image modality is used as a feature engineering method through a CNN algorithm.

[0097] Step 400, sampling the crop fullness rate of the chickens every hour, and when the crop fullness rate of the sampled chickens is lower than 95%, regenerating the feeding strategy for each group;

[0098] Step 500, adjust the grouping of the chicken flock every other day.

[0099] In one embodiment of the present invention, pellet feed is used for white-feathered broilers at different ages, and the feed corresponding to the specific age is as follows:

[0100] (1) Feed broiler chicks 0-6 days old with crushed pellet feed, small crushing;

[0101] (2) Feeding broiler chicks from 7 to 13 days old with crushed pellet feed, large crushing;

[0102] (3) Feeding broiler chickens pellet feed at 14-27 days of age;

[0103] (4) 28 days old - Feed broiler chickens with pellet feed before delivery.

[0104] In one embodiment of the present invention, the third encoder is connected to the second fully connected layer for independent pre-training, and the training calculation formula is as follows:

[0105] ;

[0106] in represents the loss value, represents the total number of training samples, represents the true leg health score annotated by the expert for the qth sample, Represents the leg health score of the qth sample output by the model.

[0107] In one embodiment of the present invention, the training of the feeding strategy model is performed by reinforcement learning, and the training steps of the feeding strategy model are as follows:

[0108] Step 101, copy the feeding strategy model to generate an Actor network and a TargetActor network, use a feedforward neural network as a Critic network, copy the Critic network to obtain a TargetCritic network, and randomly initialize the parameters of the Actor network, the TargetActor network, the Critic network, and the TargetCritic network;

[0109] Step 102, input The recorded information collected Go to the Actor network to get the Feeding strategy implemented ;

[0110] No. Feeding strategy implemented Add random noise to get Noise-enhanced feeding strategy ;

[0111] The first Feeding strategy implemented Input into the environment and get immediate rewards from feedback and The recorded information collected ; Experience , , , Deposit into experience pool;

[0112] Step 103, randomly obtain N experiences from the experience pool;

[0113] enter Go to the TargetActor network to get the Feeding strategy implemented ;

[0114] Step 104, calculate the first loss value and the second loss value , gradient descent method updates the Actor network and Critic network;

[0115] ;

[0116] ;

[0117] ;

[0118] in Represents input and The output of the Critic network is Indicates execution The comprehensive Q value when express and The output of the TargetCritic network;

[0119] ;

[0120] in, express The crop fullness rate of the chickens at all times, represents the target crop fullness rate, express The number of healthy chickens at all times, represents the total number of chickens, express The cost of the strategy adopted at any time, represents the total cost, Indicates immediate reward;

[0121] is the discount factor, which is a value between 0 and 1, used to balance the weights of immediate rewards and future rewards. The default value is 0.6;

[0122] pass Update the Actor network by Update the Critic network;

[0123] Step 105, every fixed number of network updates, update the TargetActor so that its parameters are the same as those of the current Actor network, and update the TargetCritic network so that its parameters are the same as those of the current Critic network;

[0124] Step 106, until the Actor network and the Critic network converge, the step is terminated, or after the number of cycles of steps 103-105 reaches a set value, the Actor network replaces the feeding strategy model and the step is terminated.

[0125] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation mode. The above-mentioned specific implementation mode is merely illustrative and not restrictive. Under the guidance of this embodiment, ordinary technicians in this field can also make more forms of equivalent embodiments, all of which are within the protection of this embodiment.

Claims

1. A method for optimizing the ecological breeding of white-feathered broiler chickens, characterized in that: The following steps are involved: Step 100, the white-feathered broiler chickens are artificially divided into three groups: a healthy group, a slightly lame group, and a severely lame group, and the record information of each group and the multi-dimensional record information of the chicken group are collected every hour, and the record information includes feeding records, chicken group information, and environmental information; The multi-dimensional recorded information of the chicken flock includes: chicken flock information, environmental information, and chicken flock image information; Step 200, generating plane data No. 1 based on the record information collected at the three most recent consecutive time points, and generating plane data No. 2 based on the collected multi-dimensional record information of the chicken flock; When the collected data is less than three times, artificial experience is used for feeding; Step 300, inputting the first plane data and the second plane data into a feeding strategy model, and outputting a feeding strategy representing each group; Step 400, sampling the crop fullness rate of the chickens every hour, and when the crop fullness rate of the sampled chickens is lower than 95%, regenerating the feeding strategy for each group; Step 500, adjust the grouping of the chicken flock every other day.

2. The method for optimizing ecological breeding of white-feathered broiler chickens according to claim 1, characterized in that: Feeding records include: protein content, fat content, calcium content, phosphorus content, vitamin content, feed amount, and feeding frequency in each group feed; Chicken flock information includes: the number, average weight and age of each group of chickens; Environmental information includes: temperature, humidity, and light; The chicken flock image information includes: high-resolution images of each group of chicken flocks.

3. The method for optimizing ecological breeding of white-feathered broilers according to claim 1, characterized in that: The first plane data includes a first bin for storing data, a serial data stream of a first bin indexing a first data object, and a first relationship list for storing association relationships between the first data objects indexed by the first bin, wherein the element value of the i-th row and j-th column of the first relationship list represents the quantitative value of the association relationship between the first data objects indexed by the i-th and j-th first bins; The association relationships include high and low, corresponding to two quantitative values; Data objects No. 1 include flocks, flock groups, environment, and feeding records; The first data objects with high association relationships are defined as: chicken flocks and their environment, chicken flocks and the flock group they belong to, and feeding records and the flocks they belong to; A serial data stream of a data object No. 1 includes n data units sorted by time, and the tth data unit represents the record information collected at the tth moment of the data object.

4. The method for optimizing ecological breeding of white-feathered broiler chickens according to claim 1, characterized in that: The second plane data includes a second facet for storing data, one second facet indexes a second data object, and the second plane data also includes a second relationship list for storing association relationships between the second data objects indexed by the second facet, and the element value of the i-th row and j-th column of the second relationship list represents the quantitative value of the association relationship between the second data objects indexed by the i-th and j-th second facets; Data objects No. 2 include chicken flocks, chicken flock groups, environments, and chicken flock images; The second data object with a high correlation relationship is defined as: a flock of chickens and its environment, a flock of chickens and the flock group to which it belongs, and a flock of chickens in a flock image including the flock of chickens in the flock image.

5. The method for optimizing ecological breeding of white-feathered broiler chickens according to claim 1, characterized in that: The feeding strategy model includes a first encoder, a second encoder, a third encoder, a first feature fusion layer, a first fully connected layer and a second fully connected layer, wherein the first encoder is used to input plane data No. 1 and output a first identification feature to the second encoder, the second encoder outputs a second identification feature to the first feature fusion layer, the third encoder inputs plane data No. 2 and outputs a third identification feature to the first feature fusion layer and the second fully connected layer, the second fully connected layer outputs a leg health status of each group, the first feature fusion layer outputs the first fusion feature to the first fully connected layer, and the first fully connected layer outputs a feeding strategy of each group.

6. The method for optimizing ecological breeding of white-feathered broiler chickens according to claim 1, characterized in that: The multi-dimensional record information of the chicken flock is input into the model after feature engineering. The multi-dimensional record information of the chicken flock includes image modality. The CNN algorithm is used as a feature engineering method for the image modality information.

7. The method for optimizing ecological breeding of white-feathered broilers according to claim 1, characterized in that: The third encoder is connected to the second fully connected layer for independent pre-training. The training calculation formula is as follows: ； in represents the loss value, represents the total number of training samples, represents the true leg health score annotated by the expert for the qth sample, Represents the leg health score of the qth sample output by the model.

8. The method for optimizing ecological breeding of white-feathered broilers according to claim 1, characterized in that: The training of the feeding strategy model is carried out using reinforcement learning. The training steps of the feeding strategy model are as follows: Step 101, copy the feeding strategy model to generate an Actor network and a TargetActor network, use a feedforward neural network as a Critic network, copy the Critic network to obtain a TargetCritic network, and randomly initialize the parameters of the Actor network, the TargetActor network, the Critic network, and the TargetCritic network; Step 102, input The recorded information collected Go to the Actor network to get the Feeding strategy implemented ; No. Feeding strategy implemented Add random noise to get Noise-enhanced feeding strategy ; The first Feeding strategy implemented Input into the environment and get immediate rewards from feedback and The recorded information collected ; Experience , , , Deposit into experience pool; Step 103, randomly obtain N experiences from the experience pool; enter Go to the TargetActor network to get the Feeding strategy implemented ; Step 104, calculate the first loss value and the second loss value , gradient descent method updates the Actor network and Critic network; ； ； ； in Represents input and The output of the Critic network is Indicates execution The comprehensive Q value when express and The output of the TargetCritic network; ； in, express The crop fullness rate of the chickens at all times, represents the target crop fullness rate, express The number of healthy chickens at all times, represents the total number of chickens, express The cost of the strategy adopted at any time, represents the total cost, Indicates immediate reward; is the discount factor, which is a value between 0 and 1, used to balance the weights of immediate rewards and future rewards. The default value is 0.6; pass Update the Actor network by Update the Critic network; Step 105, every fixed number of network updates, update the TargetActor so that its parameters are the same as those of the current Actor network, and update the TargetCritic network so that its parameters are the same as those of the current Critic network; Step 106, until the Actor network and the Critic network converge, the step is terminated, or after the number of cycles of steps 103-105 reaches a set value, the Actor network replaces the feeding strategy model and the step is terminated.

9. A white-feather broiler ecological breeding optimization system, characterized in that: It is used to implement the white-feather broiler ecological breeding optimization method described in any one of claims 1-8.

Images