Egg-laying hen feeding scheme adjustment method and system based on feather replacement image recognition

Abstract

The application provides an egg laying hen feeding scheme adjustment method and system based on molting image recognition, and relates to the technical field of intelligent breeding, comprising: acquiring a plurality of multi-angle body surface image sequences of a plurality of individual egg laying hens in a target chicken population synchronously collected under natural light conditions; performing semantic segmentation according to the multi-angle body surface image sequences to obtain a feather contour segmentation atlas set; performing group feature joint extraction according to the feather contour segmentation atlas set to obtain a multi-dimensional feature tensor simultaneously representing individual characteristics and group distribution information; performing group state collaborative inference according to the multi-dimensional feature tensor to obtain a three-dimensional probability distribution vector; performing group nutrition demand mapping according to the three-dimensional probability distribution vector to obtain an optimal nutrition parameter combination suitable for the entire chicken population; and performing feeding scheme generation according to the optimal nutrition parameter combination to obtain a feeding instruction stream. The application effectively improves the accuracy and adaptability of nutrition regulation during the molting period.

Description

Technical Field

[0001] This invention relates to the field of intelligent aquaculture technology, and more specifically, to a method and system for adjusting the feeding program of laying hens based on molting image recognition. Background Technology

[0002] In layer hen farming, nutritional regulation during molting is a crucial factor affecting production performance. Traditional methods primarily rely on manual observation of feather loss in laying hens to determine the molting stage, combined with farming experience, to adjust the nutritional formula. Current technology attempts to acquire surface features of the chickens using image acquisition equipment, extract single indicators such as feather coverage using conventional image processing algorithms, and then match a nutritional plan according to preset rules.

[0003] However, this method has significant limitations: first, it relies on human experience to judge the progress of molting in a group, making it difficult to accurately quantify group synchronicity; second, it ignores the dynamic impact of individual differences on the nutritional needs of the group; and third, it cannot establish a precise mapping relationship between molting stages and nutritional requirements, leading to a mismatch between nutritional supply and physiological needs. These problems directly affect molting quality and egg production recovery efficiency, hindering the improvement of refined farming practices.

[0004] Based on the shortcomings of the existing technology, there is an urgent need for a method and system for adjusting the feeding program of laying hens based on molting image recognition. Summary of the Invention

[0005] The purpose of this invention is to provide a method for adjusting the feeding program of laying hens based on molting image recognition, so as to improve the above-mentioned problems. To achieve the above objective, the technical solution adopted by this invention is as follows:

[0006] In a first aspect, this application provides a method for adjusting the feeding program of laying hens based on molting image recognition, including:

[0007] A sequence of multi-angle body surface images of multiple laying hens in a target flock was acquired synchronously under natural light conditions. The sequence of multi-angle body surface images includes the spatial distribution relationship and temporal information of individuals in the flock.

[0008] Semantic segmentation is performed on the multi-angle body surface image sequence to obtain a feather contour segmentation map set;

[0009] Based on the feather contour segmentation map set, joint extraction of group features is performed. By constructing a similarity measurement matrix between individuals in the feature space, a multidimensional feature tensor that simultaneously represents individual characteristics and group distribution information is obtained.

[0010] Based on the multidimensional feature tensor, collaborative inference of the group state is performed. By aggregating individual features and constructing a group state probability model, a three-dimensional probability distribution vector representing the uniform progress of the entire flock's molting stage is obtained.

[0011] Based on the three-dimensional probability distribution vector, the nutritional requirements of the flock are mapped, and the nutrient ratio constraints of the flock are solved by constructing a multi-objective optimization function to obtain the optimal combination of nutritional parameters applicable to the entire flock.

[0012] Feeding plans are generated based on the optimal combination of nutritional parameters to obtain a feeding instruction flow applicable to the entire flock during continuous molting.

[0013] Secondly, this application also provides a system for adjusting the feeding program of laying hens based on molting image recognition, including:

[0014] The acquisition module is used to acquire a sequence of multi-angle body surface images of multiple laying hens in a target flock under natural light conditions. The sequence of multi-angle body surface images includes the spatial distribution relationship and temporal information between individuals in the flock.

[0015] The segmentation module is used to perform semantic segmentation based on the multi-angle body surface image sequence to obtain a set of feather contour segmentation maps.

[0016] The extraction module is used to perform joint extraction of group features based on the feather contour segmentation map set. By constructing a similarity measurement matrix between individuals in the feature space, a multidimensional feature tensor that simultaneously represents individual characteristics and group distribution information is obtained.

[0017] The inference module is used to perform collaborative inference of the group state based on the multidimensional feature tensor. By aggregating individual features and constructing a group state probability model, a three-dimensional probability distribution vector representing the uniform progress of the entire flock's molting stage is obtained.

[0018] The mapping module is used to map the nutritional requirements of the flock based on the three-dimensional probability distribution vector, and solve the nutrient ratio constraints of the flock by constructing a multi-objective optimization function to obtain the optimal combination of nutritional parameters applicable to the entire flock.

[0019] The generation module is used to generate a feeding plan based on the optimal combination of nutritional parameters, resulting in a feeding instruction flow applicable to the entire flock during the continuous molting stage.

[0020] The beneficial effects of this invention are as follows:

[0021] This invention achieves accurate identification and dynamic matching of nutritional needs during the molting stage of chicken flocks by combining group collaborative semantic segmentation and feature extraction based on multi-angle image sequences with group state collaborative inference and nutritional dynamics modeling. This enables the automatic generation of precise feeding programs that conform to the physiological characteristics of the flock, effectively improving the accuracy and adaptability of nutritional regulation during the molting period. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a schematic diagram of a method for adjusting the feeding program of laying hens based on molting image recognition, as described in an embodiment of the present invention.

[0024] Figure 2 This is a schematic diagram of a system for adjusting the feeding scheme of laying hens based on molting image recognition, as described in an embodiment of the present invention.

[0025] Figure 3 This is a schematic diagram of the structure of an equipment for adjusting a feeding scheme for laying hens based on molting image recognition, as described in an embodiment of the present invention.

[0026] The diagram is labeled as follows: 800, a device for adjusting the feeding program of laying hens based on molting image recognition; 801, processor; 802, memory; 803, multimedia component; 804, I / O interface; 805, communication component; 901, acquisition module; 902, segmentation module; 903, extraction module; 904, inference module; 905, mapping module; 906, generation module. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0028] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0029] Example 1:

[0030] This embodiment provides a method for adjusting the feeding program of laying hens based on molting image recognition.

[0031] See Figure 1 The figure shows that the method includes steps S100 to S600.

[0032] Step S100: Obtain a sequence of multi-angle body surface images of multiple laying hens in the target flock under natural light conditions. The multi-angle body surface image sequence includes the spatial distribution relationship and temporal information of individuals in the flock.

[0033] Understandably, this step employs a spatially stratified random sampling strategy, uniformly selecting representative sampling points within the chicken house to ensure coverage of individual laying hens at different cage levels and in different areas. This method is based on the biological characteristics of individual asynchrony and group convergence during molting in laying hens. By simultaneously triggering multiple image acquisition devices, high-resolution image sequences of the back, abdomen, and sides of the flock are captured under natural lighting conditions. This not only records the details of the feather morphology of individual individuals but also preserves the relative positional relationships and temporal correlations of behaviors between individuals through timestamps and spatial coordinate markers. This design fully considers the three-dimensionality of poultry feather growth, regional differences, and the mutual influence of group behavior, providing a complete data foundation including spatiotemporal correlation characteristics for subsequent group collaborative analysis and laying the foundation for an overall solution approach to inferring group status from individual characteristics.

[0034] Step S200: Perform semantic segmentation based on the multi-angle body surface image sequence to obtain a set of feather contour segmentation maps;

[0035] It should be noted that this step uses semantic segmentation technology to process multi-angle image sequences. Taking into account the complex texture transition between bird feathers and skin background, deep learning algorithms are used to accurately separate the feather region, forming a consistent set of segmentation maps, providing standardized input for group feature analysis.

[0036] Step S300: Perform joint extraction of group features based on the feather contour segmentation map set. By constructing a similarity measurement matrix between individuals in the feature space, a multidimensional feature tensor that simultaneously represents individual characteristics and group distribution information is obtained.

[0037] Understandably, this step, by constructing a similarity metric matrix in the feature space, fully considers the interrelationship of feather growth status among individuals in the laying hen population, integrates individual characteristics with the population distribution pattern, and forms a multidimensional feature expression that can reflect both individual differences and population patterns.

[0038] Step S400: Perform collaborative inference of the group state based on the multidimensional feature tensor. By aggregating individual features and constructing a group state probability model, a three-dimensional probability distribution vector representing the uniform progress of the entire flock's molting stage is obtained.

[0039] It should be noted that this step is based on multidimensional feature tensors for collaborative inference of the group state. This method addresses the characteristic of the individual progress not being completely synchronized during the molting process of laying hens. It uses a probability model to quantify the overall progress state of the group, accurately captures the transition characteristics of the molting stage, and provides a precise time-series judgment basis for nutritional regulation.

[0040] Step S500: Map the nutritional requirements of the flock based on the three-dimensional probability distribution vector, solve the nutrient ratio constraints of the flock by constructing a multi-objective optimization function, and obtain the optimal combination of nutritional parameters applicable to the entire flock.

[0041] Understandably, this step solves the nutrient ratio constraints through a multi-objective optimization function, taking into account the dynamic changes in the requirements of laying hens for nutrients such as protein, amino acids and minerals during molting, establishing a precise mapping relationship between the nutritional requirements of the flock and the molting stage, and achieving refined regulation of nutrient supply.

[0042] Step S600: Generate a feeding plan based on the optimal combination of nutritional parameters to obtain a feeding instruction flow applicable to the entire flock during the continuous molting stage.

[0043] It should be noted that this step generates a feeding instruction flow based on optimal nutritional parameters. Considering the continuous and gradual nature of the molting process, the feeding plan is dynamically adjusted to ensure that the nutritional supply matches the needs of feather growth, thus realizing the conversion from nutritional parameters to actual feeding operations.

[0044] Further, step S200 includes steps S210 to S230.

[0045] Step S210: Perform initial semantic segmentation processing based on the multi-angle body surface image sequence. By introducing the multi-scale feature extraction module of the adversarial generative network, perform pixel-level prediction on a single sample to obtain the initial feather region probability map.

[0046] Step S220: Perform spatial consistency optimization processing based on the initial feather region probability map. By constructing a group spatial relationship model based on graph attention mechanism, the spatial distribution relationship between individuals is used to constrain the group consistency of the segmentation boundary, and a spatially optimized segmentation map is obtained.

[0047] Step S230: Perform temporal consistency fusion processing based on the spatial optimization segmentation map. By establishing a temporal regularization constraint term, the temporal information of multi-angle images is fused to perform feature alignment across time periods, resulting in the final feather contour segmentation map set.

[0048] Specifically, step S210 first uses the multi-scale feature extraction module of an adversarial generative network to perform pixel-level semantic segmentation on a single laying hen image. This network, through an adversarial training mechanism between the generator and the discriminator, is particularly optimized for recognizing blurred regions at the junction of feathers and skin, generating an initial feather region probability map. Based on this, step S220 introduces a group spatial relationship model with a graph attention mechanism, transforming the spatial distribution relationship of individuals in the flock into node connection weights in a graph structure. The segmentation boundary constraint between adjacent individuals is dynamically adjusted through the attention coefficient, ensuring that individuals in the same molting stage in the flock maintain the consistency of the segmentation boundary. Step S230 further establishes a temporal regularization constraint term, dynamically warping the multi-angle image features collected at different time points. By minimizing the temporal difference loss function, feature alignment across time periods is achieved, ultimately forming a set of feather contour segmentation maps with spatiotemporal consistency.

[0049] Further, step S300 includes steps S310 to S330.

[0050] Step S310: Perform individual feature encoding processing based on the feather contour segmentation map set. By constructing a spatial-texture feature extractor based on a three-dimensional convolutional neural network, the spatial distribution morphology and texture features of the feathers are analyzed simultaneously to obtain the initial individual feature vector set.

[0051] Step S320: Perform group similarity modeling based on the initial individual feature vector set. By designing a dual measurement mechanism based on cosine similarity and Euclidean distance, construct a group similarity matrix that represents the feature correlation between individuals.

[0052] Step S330: Perform feature fusion processing based on the group similarity matrix. By introducing a feature aggregation network based on attention weights, the individual feature vectors and the group similarity matrix are nonlinearly fused to obtain a multidimensional feature tensor that simultaneously represents individual characteristics and group distribution information.

[0053] Specifically, step S310 first employs a three-dimensional convolutional neural network to construct a spatial-texture feature extractor. This network simultaneously processes the spatial and texture dimensions of the feather image through three-dimensional convolutional kernels, effectively capturing the three-dimensional distribution features of feather growth and surface micro-texture changes, generating an initial individual feature vector set containing fine features such as feather root density, new feather angle, and coverage area; the extraction formula is expressed as:

[0054] ;

[0055] in, Indicates the first stage of the molting process In each time frame, the region located at position (x, y) in the chicken body surface image is the feature intensity value of a specific texture or shape identified by the m-th feature detector (convolution kernel); This represents the weighting coefficient used in the m-th feature detector to multiply the input pixel value located at time offset δ and spatial position (Φ, ω) in the c-th color channel; Indicates the first video sequence in the input video sequence. In a frame image, the brightness value of the c-th color channel of a pixel located at spatial position (x+Φ, y+ω); D represents the temporal depth of the convolution kernel; F represents the spatial size of the convolution kernel; M represents the number of output feature maps; The output time coordinate is represented by x; the output height coordinate by y; the output width coordinate by m; the feature detector number by m; the time dimension offset by δ; the height offset by Φ; the width offset by ω; and the input channel index by c. m This represents the bias term of the m-th feature map, used to adjust the baseline of the output feature values.

[0056] Building upon this, step S320 innovatively employs a dual measurement mechanism of cosine similarity and Euclidean distance. Cosine similarity focuses on evaluating the directional differences in feather texture patterns, while Euclidean distance quantifies the absolute differences in spatial distribution characteristics. A dual-weight fusion algorithm is used to construct a group similarity matrix that accurately reflects the similarity of molting progress among individuals. The quantification process is expressed as follows:

[0057] ;

[0058] ;

[0059] ;

[0060] in, Cosine similarity is used to measure the directional similarity of feature vectors between two individuals. represents the Euclidean distance, used to measure the absolute distance between the feature vectors of two individuals; p and q represent the indices of the laying hens; l represents the feature dimension index; L represents the total number of feature dimensions; This represents the l-th eigenvalue of individual p; This represents the l-th eigenvalue of individual q; α represents the overall similarity score between individuals p and q; α represents the cosine similarity weight coefficient; β represents the distance similarity weight coefficient; γ represents the distance scaling coefficient; and exp represents the exponential function.

[0061] Step S330 further introduces an attention-weighted feature aggregation network. This network assigns dynamic weights to each individual feature vector through an adaptive learning mechanism, focusing on strengthening the contribution of individuals with significant molting characteristics. Simultaneously, it uses the group similarity matrix as a constraint for nonlinear feature fusion, ultimately generating a multidimensional feature tensor that retains both individual feather growth details and reflects the group's distribution patterns. These three steps, progressing layer by layer, from fine feature extraction of individual laying hens to group similarity modeling and feature fusion, effectively solve the feature discretization problem caused by individual physiological differences during poultry molting, providing a feature representation that combines individual specificity and group consistency for subsequent group status judgment. The feature fusion process is represented as follows:

[0062] ;

[0063] ;

[0064] ;

[0065] in, This represents the attention weight of individual p to individual q; The index represents the summation index, used to calculate the softmax normalization; N represents the total number of sums. This represents the overall similarity score between individuals p and r; The l-th dimension of the feature matrix represents individual q; The l-th dimension of the intermediate feature matrix represents individual p; Represents the relationship from input dimension l to output dimension l Transformation weights; Represents the first individual p The output feature matrix is ​​dimensional; σ represents the nonlinear activation function.

[0066] Further, step S400 includes steps S410 to S430.

[0067] Step S410: Perform feather state quantization processing based on multidimensional feature tensor. By constructing a nonlinear mapping function based on feather growth dynamics, quantify feather root visibility, new feather ratio and feather coverage density feature indicators to obtain individual feather state quantization vector.

[0068] Step S420: Perform group synchronization analysis based on the individual feather state quantization vector. By establishing a group state alignment model based on the dynamic time warping algorithm, analyze the time offset and synchronization degree of the molting process between individuals to obtain the group molting synchronization matrix.

[0069] Step S430: Perform cooperative state inference processing based on the group molting synchronization matrix. By introducing a group state transition mechanism based on a hidden Markov model, individual quantitative features and group synchronization information are integrated to obtain a three-dimensional probability distribution vector representing the uniform progress of the entire flock's molting stage.

[0070] Specifically, step S410 first constructs a nonlinear mapping function based on feather growth dynamics. This function is specifically designed for the biological characteristics of feather root exposure, new feather growth, and changes in feather coverage during the molting process of laying hens. Through nonlinear transformation, the multidimensional feature tensor is converted into quantifiable feather root visibility index, new feather proportion coefficient, and feather coverage density index, forming a state quantification vector that accurately represents the progress of individual molting. Step S420 then uses a dynamic time warping algorithm to establish a group state alignment model. This algorithm can effectively handle the time asynchrony problem of different individuals' molting processes. By finding the optimal matching path of the molting state sequence between individuals, the time offset and synchronization coefficient are calculated, and a synchronization matrix reflecting the degree of coordination of group molting is constructed. Step S430 finally introduces the group state transition mechanism of the Hidden Markov Model. Individual quantification features are used as observation sequences, and the group synchronization matrix is ​​used as state transition constraints. The probability distribution of the group at different molting stages (start, peak, and end) is derived through a forward-backward algorithm, and finally a three-dimensional probability distribution vector that reflects both individual states and the group's co-evolutionary laws is generated. These three steps—from precise quantification of individual states to group temporal alignment and then to collaborative state inference—effectively solve the problem of asynchronous progress caused by individual physiological differences during molting in laying hens, providing an accurate and reliable basis for judging the group state for subsequent nutritional regulation.

[0071] Further, step S500 includes steps S510 to S530.

[0072] Step S510: Perform population nutrient dynamics modeling based on the three-dimensional probability distribution vector. By introducing a multi-compartment model from pharmacokinetics, simulate the absorption, distribution, and metabolism of nutrients in the chicken population, establish the dynamic response relationship between the molting stage and nutritional requirements, and obtain the population's basic nutrient requirements spectrum.

[0073] Understandably, this step introduces a multi-compartment model from pharmacokinetics to simulate the dynamic processes of nutrients within a chicken flock. This model treats the flock as a holistic system, modeling the absorption, distribution, and metabolism of nutrients such as proteins, amino acids, and minerals as different inter-compartmental material transfer processes. By establishing a dynamic response function between the molting stage and nutritional requirements, it accurately describes the changing patterns of nutrient requirements at different molting periods. The choice to introduce the multi-compartment model from pharmacokinetics into the field of laying hen nutritional requirement modeling is based on the following considerations: First, the absorption, distribution, and metabolism of nutrients in laying hens have highly similar dynamic characteristics to the kinetic processes of drugs in organisms, both requiring description of the migration, transformation, and utilization rates of substances within the system; second, the molting process is essentially a time function of a physiological state, exhibiting a mathematical isomorphism with the time-kinetic characteristics of drug metabolism. In practical applications, three key improvements were made to the classic multi-compartment model: First, the concepts of "central compartment" and "peripheral compartment" in the traditional model were redefined as "directly available nutrient pool" and "physiological reserve nutrient pool" to more accurately describe the competitive distribution of nutrients among feather growth and other physiological functions. Second, molting-specific parameters were introduced, transforming the three-dimensional probability distribution vector into a regulator of intercompartmental transfer rates, enabling the model to dynamically respond to changes in metabolic characteristics at different molting stages. Third, a feedback mechanism based on feather protein synthesis rates was established to adjust the material flow distribution between compartments in real time by monitoring the growth status of new feathers. These improvements allow the model to accurately simulate the changing patterns of nutrient requirements during molting, ultimately outputting a population basal nutrient requirement spectrum with time resolution, providing a theoretical basis for subsequent precise nutrient regulation.

[0074] Step S520: Based on the population's basic nutritional needs spectrum, construct a multi-objective optimization function. By establishing a multi-objective constraint system that includes protein utilization, amino acid balance, and mineral synergistic effects, construct a population nutrient ratio optimization function based on Pareto optimal solution.

[0075] It should be noted that this step constructs a multi-objective optimization function based on the population's basic nutritional needs spectrum. This function simultaneously considers three mutually restrictive optimization objectives: maximizing protein utilization, optimizing amino acid balance, and maximizing the synergistic effect of minerals. By establishing a Pareto optimal solution search mechanism, the optimal ratio of each nutrient is found under the premise of meeting the population's nutritional needs, thus forming a nutrient ratio optimization function that can balance multiple optimization objectives.

[0076] Step S530: Perform dynamic optimization solution processing based on the population nutrient ratio optimization function. By simulating the dynamic distribution process of nutrients in the population, obtain the optimal combination of nutrient parameters applicable to the entire molting stage of the chicken flock.

[0077] Specifically, this step uses a dynamic response surface model for optimization. This algorithm simulates the dynamic distribution process of nutrients in the population, iteratively optimizes the ratio of each nutrient, and finally obtains a set of optimal parameter combinations that can meet the nutritional needs of the population during the molting stage and achieve efficient utilization of nutrients, providing a scientific basis for precision feeding.

[0078] Further, step S600 includes steps S610 to S630.

[0079] Step S610: Generate dynamic feeding curves based on the optimal combination of nutritional parameters. By constructing a nutritional requirement change function with the molting time series as the independent variable, a time-based baseline feeding program curve is obtained.

[0080] Step S620: Perform real-time feedback adjustment processing based on the baseline feeding program curve. By establishing an adaptive correction algorithm based on group feeding behavior monitoring data, the feeding curve is dynamically calibrated to obtain an optimized dynamic feeding program.

[0081] Step S630: Perform instruction stream serialization processing according to the dynamic feeding plan. Convert the continuous feeding curve into an instruction sequence that can be executed by the feeding equipment through the time discretization method to obtain a precise feeding instruction stream applicable to the entire flock's continuous molting stage.

[0082] Step S610 first constructs a nutrient requirement variation function with the molting time series as the independent variable. This function is based on the ratio of each nutrient in the optimal nutrient parameter combination. A nonlinear regression method is used to establish a mapping model between the number of molting days and the nutrient requirement. By introducing a time decay factor and a growth acceleration factor, the dynamic change law of nutrient requirement during molting is simulated, generating a baseline feeding program curve that shows an upward trend followed by a downward trend. Step S620, based on this, establishes an adaptive correction algorithm based on the flock's feeding behavior monitoring data. By collecting data on the flock's feeding speed, feed intake, and remaining feed troughs in real time, the deviation between the flock's actual intake and theoretical requirements is calculated. A fuzzy control algorithm is used to calibrate the baseline curve in real time, and dynamic adjustments are made specifically for the flock's appetite fluctuation characteristics during molting. Step S630, finally, a time discretization method is used to convert the continuous nutrient requirement curve into discrete feeding instructions. By setting time slices and nutrient concentration gradients, the smooth curve is converted into a sequence of instructions that the feeding equipment can recognize, ensuring that the nutrient supply in each time period is accurately matched with the flock's molting progress. These three steps—from generating theoretical curves to real-time feedback optimization and then to converting equipment instructions—effectively solve the problem of matching the dynamic changes in nutritional requirements with feeding execution during molting, achieving a seamless connection between nutritional parameters and actual feeding operations.

[0083] Example 2:

[0084] like Figure 2 As shown, this embodiment provides a system for adjusting the feeding program of laying hens based on molting image recognition. The system includes:

[0085] The acquisition module 901 is used to acquire a sequence of multi-angle body surface images of multiple laying hens in the target flock under natural light conditions. The sequence of multi-angle body surface images includes the spatial distribution relationship and temporal information of individuals in the flock.

[0086] The segmentation module 902 is used to perform semantic segmentation based on multi-angle body surface image sequences to obtain a set of feather contour segmentation maps.

[0087] The extraction module 903 is used to jointly extract group features based on the feather contour segmentation map set. By constructing a similarity measurement matrix between individuals in the feature space, a multidimensional feature tensor that simultaneously represents individual characteristics and group distribution information is obtained.

[0088] The inference module 904 is used to perform collaborative inference of the group state based on the multidimensional feature tensor. By aggregating individual features and constructing a group state probability model, a three-dimensional probability distribution vector representing the uniform progress of the entire flock's molting stage is obtained.

[0089] The mapping module 905 is used to map the nutritional requirements of the flock based on the three-dimensional probability distribution vector. By constructing a multi-objective optimization function, the group nutrient ratio constraints are solved to obtain the optimal combination of nutritional parameters applicable to the entire flock.

[0090] The generation module 906 is used to generate a feeding program based on the optimal combination of nutritional parameters, resulting in a feeding instruction flow applicable to the entire flock during the continuous molting stage.

[0091] In one specific embodiment of this application, the segmentation module 902 includes:

[0092] The first segmentation unit is used to perform initial semantic segmentation processing based on the multi-angle body surface image sequence. By introducing the multi-scale feature extraction module of the adversarial generative network, pixel-level prediction is performed on a single sample to obtain the initial feather region probability map.

[0093] The second segmentation unit is used to perform spatial consistency optimization processing based on the initial feather region probability map. By constructing a group spatial relationship model based on graph attention mechanism, the spatial distribution relationship between individuals is used to constrain the group consistency of the segmentation boundary, and a spatially optimized segmentation map is obtained.

[0094] The third segmentation unit is used to perform temporal consistency fusion processing based on the spatially optimized segmentation map. By establishing a temporal regularization constraint term, it fuses the temporal information of multi-angle images to perform feature alignment across time periods, thereby obtaining the final feather contour segmentation map set.

[0095] In one specific embodiment of this application, the extraction module 903 includes:

[0096] The first extraction unit is used to encode individual features based on the feather contour segmentation map set. By constructing a spatial-texture feature extractor based on a three-dimensional convolutional neural network, the spatial distribution morphology and texture features of the feathers are simultaneously analyzed to obtain an initial set of individual feature vectors.

[0097] The second extraction unit is used to perform group similarity modeling based on the initial individual feature vector set. By designing a dual measurement mechanism based on cosine similarity and Euclidean distance, a group similarity matrix representing the feature correlation between individuals is constructed.

[0098] The third extraction unit is used to perform feature fusion processing based on the group similarity matrix. By introducing a feature aggregation network based on attention weights, the individual feature vectors and the group similarity matrix are nonlinearly fused to obtain a multidimensional feature tensor that simultaneously represents individual characteristics and group distribution information.

[0099] In one specific embodiment of this application, the inference module 904 includes:

[0100] The first inference unit is used to quantify the feather state based on the multidimensional feature tensor. By constructing a nonlinear mapping function based on feather growth dynamics, it quantifies the feather root visibility, the proportion of new feathers and the feather coverage density feature index to obtain the individual feather state quantification vector.

[0101] The second inference unit is used to perform group synchronization analysis based on the individual feather state quantization vector. By establishing a group state alignment model based on the dynamic time warping algorithm, the time offset and synchronization degree of the molting process between individuals are analyzed to obtain the group molting synchronization matrix.

[0102] The third inference unit is used to perform cooperative state inference processing based on the group molting synchronization matrix. By introducing a group state transition mechanism based on a hidden Markov model, it integrates individual quantitative features and group synchronization information to obtain a three-dimensional probability distribution vector that represents the uniform progress of the entire flock's molting stage.

[0103] In one specific embodiment of this application, the mapping module 905 includes:

[0104] The first mapping unit is used to perform population nutrient dynamics modeling based on the three-dimensional probability distribution vector. By introducing a multi-compartment model from pharmacokinetics, it simulates the absorption, distribution, and metabolism of nutrients in the chicken population, establishes the dynamic response relationship between the molting stage and nutritional requirements, and obtains the population's basic nutritional requirements spectrum.

[0105] The second mapping unit is used to construct a multi-objective optimization function based on the population's basic nutritional needs spectrum. By establishing a multi-objective constraint system that includes protein utilization, amino acid balance, and mineral synergistic effects, a population nutrient ratio optimization function based on Pareto optimal solution is constructed.

[0106] The third mapping unit is used to perform dynamic optimization based on the population nutrient ratio optimization function. By simulating the dynamic distribution process of nutrients in the population, it obtains the optimal combination of nutrient parameters applicable to the entire molting stage of the flock.

[0107] Example 3:

[0108] Corresponding to the above method embodiments, this embodiment also provides a layer hen feeding program adjustment device based on molting image recognition. The layer hen feeding program adjustment device based on molting image recognition described below and the layer hen feeding program adjustment method based on molting image recognition described above can be referred to in correspondence.

[0109] Figure 3 This is a block diagram illustrating an egg-laying hen feeding program adjustment device 800 based on molting image recognition, according to an exemplary embodiment. Figure 3 As shown, the layer hen feeding program adjustment device 800 based on molting image recognition may include: a processor 801 and a memory 802. The layer hen feeding program adjustment device 800 based on molting image recognition may also include one or more of the following: a multimedia component 803, an I / O interface 804, and a communication component 805.

[0110] The processor 801 controls the overall operation of the molting image recognition-based layer hen feeding program adjustment device 800 to complete all or part of the steps in the aforementioned molting image recognition-based layer hen feeding program adjustment method. The memory 802 stores various types of data to support the operation of the molting image recognition-based layer hen feeding program adjustment device 800. This data may include, for example, instructions for any application or method operating on the molting image recognition-based layer hen feeding program adjustment device 800, as well as application-related data such as contact data, sent and received messages, images, audio, video, etc. The memory 802 can be implemented using any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The multimedia component 803 may include a screen and an audio component. The screen may be, for example, a touchscreen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signals may be further stored in the memory 802 or transmitted via the communication component 805. The audio component also includes at least one speaker for outputting audio signals. I / O interface 804 provides an interface between processor 801 and other interface modules, such as keyboards, mice, and buttons. These buttons can be virtual or physical. Communication component 805 is used for wired or wireless communication between the molting image recognition-based layer hen feeding scheme adjustment device 800 and other devices. Wireless communication includes Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, or 4G, or a combination thereof. Therefore, the corresponding communication component 805 may include a Wi-Fi module, a Bluetooth module, and an NFC module.

[0111] In an exemplary embodiment, a layer hen feeding program adjustment device 800 based on molting image recognition may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the aforementioned layer hen feeding program adjustment method based on molting image recognition.

[0112] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided. When executed by a processor, these program instructions implement the steps of the above-described method for adjusting a layer hen feeding program based on molting image recognition. For example, the computer-readable storage medium may be the memory 802 including the program instructions, which may be executed by a processor 801 of a layer hen feeding program adjustment device 800 based on molting image recognition to complete the above-described method for adjusting a layer hen feeding program based on molting image recognition.

[0113] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A laying hen feeding program adjustment method based on feather exchange image recognition, characterized in that, The method comprises the following steps: acquiring a plurality of multi-angle body surface image sequences of a plurality of individual hens in a target flock under natural light conditions, the multi-angle body surface image sequences comprising spatial distribution relationships and time sequence information among the individuals in the flock; performing semantic segmentation on the multi-angle body surface image sequences to obtain a set of feather contour segmentation atlases; performing joint extraction of flock characteristics according to the set of feather contour segmentation atlases, constructing a similarity measurement matrix among the individuals in a feature space to obtain a multi-dimensional feature tensor representing individual characteristics and flock distribution information; performing collaborative inference of flock state according to the multi-dimensional feature tensor, aggregating individual features and constructing a flock state probability model to obtain a three-dimensional probability distribution vector representing the unified progress of the molting stage of the entire flock; performing mapping of flock nutritional requirements according to the three-dimensional probability distribution vector, constructing a multi-objective optimization function to solve the constraint conditions of flock nutrient ratio to obtain an optimal nutrient parameter combination suitable for the entire flock; generating a feeding scheme according to the optimal nutrient parameter combination to obtain a feeding instruction stream suitable for the entire flock during the continuous molting stage; wherein the collaborative inference of flock state according to the multi-dimensional feature tensor, the aggregation of individual features and the construction of a flock state probability model to obtain a three-dimensional probability distribution vector representing the unified progress of the molting stage of the entire flock comprises: performing feather state quantization processing according to the multi-dimensional feature tensor, constructing a nonlinear mapping function based on feather growth dynamics to quantify feather visibility, proportion of newly grown feathers and feather coverage density characteristic indicators to obtain an individual feather state quantization vector; performing flock synchrony analysis processing according to the individual feather state quantization vector, establishing a flock state alignment model based on a dynamic time warping algorithm to analyze the time offset and synchronization degree of the molting progress among the individuals to obtain a flock molting synchrony matrix; performing collaborative state inference processing according to the flock molting synchrony matrix, introducing a flock state transition mechanism based on a hidden Markov model to fuse individual quantization features and flock synchrony information to obtain a three-dimensional probability distribution vector representing the unified progress of the molting stage of the entire flock; wherein the mapping of flock nutritional requirements according to the three-dimensional probability distribution vector, the construction of a multi-objective optimization function to solve the constraint conditions of flock nutrient ratio to obtain an optimal nutrient parameter combination suitable for the entire flock comprises: performing flock nutritional dynamics modeling processing according to the three-dimensional probability distribution vector, introducing a multi-compartment model in pharmacokinetics to simulate the absorption, distribution and metabolism of nutrients in the flock to establish a dynamic response relationship between the molting stage and nutritional requirements, and obtaining a flock basic nutritional requirement spectrum; performing multi-objective optimization function construction processing according to the flock basic nutritional requirement spectrum, establishing a multi-objective constraint system including protein utilization rate, amino acid balance and mineral synergistic effect, and constructing a flock nutrient ratio optimization function based on Pareto optimal solution; According to the population nutrient ratio optimization function, dynamic optimization solving processing is performed, the dynamic allocation process of nutrients in the population is simulated, and the optimal nutrition parameter combination suitable for the whole chicken population molting stage is obtained.

2. The laying hen feeding program adjustment method based on feather exchange image recognition according to claim 1, characterized in that, According to the multi-angle body surface image sequence, semantic segmentation is performed to obtain a feather contour segmentation atlas set, including: According to the multi-angle body surface image sequence, initial semantic segmentation processing is performed, a multi-scale feature extraction module of a generative adversarial network is introduced to perform pixel-level prediction on a single sample, and an initial feather region probability map is obtained. According to the initial feather region probability map, spatial consistency optimization processing is performed, a group spatial relationship model based on a graph attention mechanism is constructed, the spatial distribution relationship between individuals is used to constrain the group consistency of the segmentation boundary, and a spatial optimization segmentation atlas is obtained. According to the spatial optimization segmentation atlas, time sequence consistency fusion processing is performed, a time sequence regularization constraint term is established, the time sequence information of multi-angle images is fused to perform cross-time period feature alignment, and a final feather contour segmentation atlas set is obtained.

3. The laying hen feeding program adjustment method based on feather exchange image recognition according to claim 1, characterized in that, According to the feather contour segmentation atlas set, group feature joint extraction is performed, a similarity measurement matrix between individuals in the feature space is constructed, a multi-dimensional feature tensor representing individual characteristics and group distribution information is obtained, including: According to the feather contour segmentation atlas set, individual feature coding processing is performed, a space-texture feature extractor based on a three-dimensional convolutional neural network is constructed, the spatial distribution form and texture features of the feather are synchronously analyzed, and an initial individual feature vector set is obtained. According to the initial individual feature vector set, group similarity modeling processing is performed, a dual measurement mechanism based on cosine similarity and Euclidean distance is designed, and a group similarity matrix representing the feature correlation between individuals is constructed. According to the group similarity matrix, feature fusion processing is performed, an attention weight-based feature aggregation network is introduced, individual feature vectors and group similarity matrices are nonlinearly fused, and a multi-dimensional feature tensor representing individual characteristics and group distribution information is obtained.

4. A laying hen feeding plan adjustment system based on feather exchange image recognition, characterized in that, Including: An acquisition module is configured to acquire a multi-angle body surface image sequence synchronously collected under natural light conditions for a plurality of hen individuals in a target chicken population, the multi-angle body surface image sequence including spatial distribution relationships and time sequence information between individuals in the population; A segmentation module is configured to perform semantic segmentation on the multi-angle body surface image sequence to obtain a feather contour segmentation atlas set; An extraction module is configured to perform group feature joint extraction on the feather contour segmentation atlas set by constructing a similarity measurement matrix between individuals in the feature space to obtain a multi-dimensional feature tensor representing individual characteristics and group distribution information; An inference module is configured to perform group state collaborative inference on the multi-dimensional feature tensor by aggregating individual features and constructing a group state probability model to obtain a three-dimensional probability distribution vector representing the unified progress of the whole chicken population molting stage; A mapping module is configured to perform group nutrition demand mapping on the three-dimensional probability distribution vector by constructing a multi-objective optimization function to solve group nutrient ratio constraint conditions to obtain an optimal nutrition parameter combination suitable for the whole chicken population. The generating module is configured to generate a feeding scheme according to the optimal combination of nutritional parameters, and obtain a feeding instruction stream suitable for a continuous molting stage of the whole flock; The inference module includes: The first inference unit is configured to perform feather state quantization processing according to the multi-dimensional feature tensor, quantize feather visibility, newly grown feather proportion and feather coverage density characteristic indexes by constructing a nonlinear mapping function based on feather growth dynamics, and obtain an individual feather state quantization vector. The second inference unit is configured to perform group synchrony analysis processing according to the individual feather state quantization vector, analyze the time offset and synchrony degree of the molting process between individuals by establishing a group state alignment model based on a dynamic time warping algorithm, and obtain a group molting synchrony matrix. The third inference unit is configured to perform collaborative state inference processing according to the group molting synchrony matrix, introduce a group state transition mechanism based on a hidden Markov model, and fuse individual quantization features and group synchrony information to obtain a three-dimensional probability distribution vector representing the unified progress of the whole flock in the molting stage. The mapping module includes: The first mapping unit is configured to perform group nutrition dynamics modeling processing according to the three-dimensional probability distribution vector, simulate the absorption, distribution and metabolism of nutrients in the chicken group by introducing a multi-compartment model in pharmacokinetics, establish a dynamic response relationship between the molting stage and nutritional requirements, and obtain a group basic nutritional requirement spectrum. The second mapping unit is configured to perform multi-objective optimization function construction processing according to the group basic nutritional requirement spectrum, establish a multi-objective constraint system including protein utilization rate, amino acid balance and mineral synergistic effect, and construct a group nutrient ratio optimization function based on Pareto optimal solution. The third mapping unit is configured to perform dynamic optimization solving processing according to the group nutrient ratio optimization function, simulate the dynamic allocation process of nutrients in the group, and obtain an optimal combination of nutritional parameters suitable for the whole flock in the molting stage.

5. The laying hen feeding regimen adjustment system based on feather image recognition according to claim 4, characterized in that, The segmentation module includes: The first segmentation unit is configured to perform initial semantic segmentation processing according to the multi-angle body surface image sequence, introduce a multi-scale feature extraction module of the generative adversarial network to perform pixel-level prediction on a single sample, and obtain an initial feather region probability map. The second segmentation unit is configured to perform spatial consistency optimization processing according to the initial feather region probability map, construct a group spatial relationship model based on a graph attention mechanism, constrain the group consistency of the segmentation boundary by using the spatial distribution relationship between individuals, and obtain a spatially optimized segmentation atlas. The third segmentation unit is configured to perform temporal consistency fusion processing according to the spatially optimized segmentation atlas, establish a temporal regularization constraint term, fuse the temporal information of multi-angle images for cross-time period feature alignment, and obtain a final set of feather contour segmentation atlases.

6. The laying hen feeding plan adjustment system based on feather image recognition according to claim 4, characterized in that, The extraction module includes: The first extraction unit is configured to perform individual feature encoding processing according to the set of feather contour segmentation atlases, construct a spatial-texture feature extractor based on a three-dimensional convolutional neural network, synchronously analyze the spatial distribution morphology and texture features of the feather, and obtain an initial individual feature vector set. The second extraction unit is configured to perform a group similarity modeling process according to the initial individual feature vector set, and construct a group similarity matrix representing the feature correlation between individuals by designing a double measurement mechanism based on cosine similarity and Euclidean distance. The third extraction unit is configured to perform a feature fusion process according to the group similarity matrix, and obtain a multi-dimensional feature tensor representing both individual characteristics and group distribution information by introducing a feature aggregation network based on attention weights to nonlinearly fuse the individual feature vector and the group similarity matrix.

Images