Breeding method and system based on multi-dimensional data

Abstract

The embodiment of the present application relates to the technical field of breeding management, and discloses a selection and breeding method based on multidimensional data, which comprises the following steps: determining the feather characteristic parameters of chicks according to feather characteristic information and pixel point size; determining the gender and predicting the growth parameter information of the corresponding chicks according to the length of the primary wing feather and the length of the over primary wing feather, and performing primary screening according to the predicted growth parameter information; inputting the image information of the breeding hen into a feature recognition model to obtain the corresponding characteristics of the breeding hen; inputting the comb image into a comb recognition model to determine the corresponding comb type; if the feather color, the chicken toe color and the comb type match the breed parameters of the corresponding breeding hen, then the breeding hen is kept and caged, and the individual characteristic parameters are recorded. The scheme of the embodiment of the present application realizes the deep fusion of image recognition technology and multidimensional phenotype data, constructs a full-cycle accurate selection and breeding system from chicks to breeding hens, and realizes the automation of gender identification, the standardization of breed verification and the intelligentization of growth prediction.

Description

Technical Field

[0001] This invention relates to the field of aquaculture management technology, specifically to a breeding method and system based on multi-dimensional data. Background Technology

[0002] In modern poultry farming, most farms still rely on traditional experience for breed selection, lacking systematic assessment and precise measurement of the flock's genetic potential. This leads to a subjective and inefficient selection process, making it difficult to effectively select breeds with excellent feed conversion ratios and high egg production. In the mating stage, farmers typically use simple group mating methods, failing to fully consider the kinship and trait complementarity among breeder chickens. This prevents the scientific selection of breeders to maximize genetic advantages, reducing the overall production performance and disease resistance of offspring. At the breeding planning level, most farms lack long-term, systematic breeding goals and plans, failing to combine market demands and industry development trends to conduct targeted breeding and improvement of important traits such as feed utilization efficiency and egg production performance, resulting in slow improvement in breed production performance.

[0003] Meanwhile, actual breeding management still generally adopts an extensive strategy, simply controlling feed intake based on the growth stage and weight of breeder chickens. It fails to effectively integrate the results of selection, mating, and breeding planning with feed input, lacks dynamic monitoring and precise control of the flock's real-time growth status, and suffers from low real-time feeding rates. This severely restricts the improvement of breeding efficiency, making it difficult to achieve expected egg production rates and hatching egg quality rates, thus limiting the sustainable development and economic benefits of the poultry industry. Therefore, designing an efficient breeding program has become a pressing technical problem for those skilled in the art. Summary of the Invention

[0004] To address the aforementioned shortcomings, this invention discloses a breeding method based on multi-dimensional data, which enables efficient breeding of breeder chickens and improves overall breeding management efficiency.

[0005] The first aspect of this invention discloses a breeding method based on multi-dimensional data, comprising:

[0006] Acquire image information of chicks in the first growth stage, wherein the chick image information includes feather image information; wherein the feather image information includes chick feather images and reference object images;

[0007] Feature recognition is performed on the feather image information to determine a reference object image in the feather image information, the pixel information of the corresponding reference object in the reference object image is determined, and the pixel size is determined based on the pixel information and the pre-stored reference object size;

[0008] After preprocessing the feather image information, it is input into the feather feature recognition model to identify the corresponding feather feature information. Based on the feather feature information and pixel size, the feather feature parameters of the chick are determined. The feather feature parameters include the length of the primary wing feathers and the length of the primary wing coverts.

[0009] The sex of the chicks and the predicted growth parameters are determined based on the length of the primary wing feathers and the length of the primary wing coverts. Primary screening is then performed based on the predicted growth parameters, and the next stage of breeding management is carried out based on the results of the primary screening.

[0010] Acquire image information of breeder chickens in the second growth stage, input the breeder chicken image information into the feature recognition model to obtain the corresponding breeder chicken feather image, foot image and comb image, and determine the feather color and foot color of the breeder chicken;

[0011] The rooster comb image is input into the rooster comb recognition model to determine the corresponding rooster comb type; if the feather color, foot color and rooster comb type match the breed parameters of the corresponding breed chicken, then the chicken is kept for breeding and put into a cage, and the corresponding individual characteristic parameters are recorded.

[0012] As an optional implementation, in a first aspect of the present invention, determining the sex of the chick and predicting growth parameters based on the primary wing feather length and the overlying primary wing feather length includes:

[0013] The mean length and the length difference are determined based on the lengths of the primary wing feathers and the overlying primary wing feathers.

[0014] The sex of the chick is determined based on the length of the primary wing feathers and the length of the primary wing coverts; if the length difference and the average length meet the set conditions, the chick is determined to be a female chick; if the length difference and the average length do not meet the set conditions, the chick is determined to be a male chick.

[0015] The first feather maturity of the corresponding female chick is determined based on the average length of the primary wing feathers, the length of the primary wing coverts, and the length of the primary wing coverts.

[0016] The maturity of the second feathers of the corresponding male chicks is determined by the average length and length of the primary wing coverts.

[0017] As an optional implementation, in the first aspect of the present invention, the step of inputting the breeder chicken image information into a feature recognition model to obtain the corresponding breeder chicken's feather image, foot image, and comb image, and determining the breeder chicken's feather color and foot color, includes:

[0018] The image information of the breeding chickens is input into the backbone network to extract convolutional feature maps, and a series of anchor frames of preset size and proportion are generated on the convolutional feature maps using a sliding window;

[0019] For each anchor box, predict the probability that it contains feathers, foot, or comb foreground, filter out low-probability regions, and fine-tune the coordinates of the foreground anchor boxes to obtain the corresponding chicken feather, foot, and comb images.

[0020] The feather image is processed using EfficientNetV2-L as the backbone network, and shallow feather features and deep feather color features are extracted to optimize the feather image region.

[0021] Convert the obtained chicken foot and comb images from RGB to HSV or Lab color space;

[0022] For each color channel in the converted color space, the lower-order moments of the corresponding image are calculated. The lower-order moments include the first-order moment, the second-order moment, and the third-order moment. The first-order moment is used to characterize the average value of the color channel, the second-order moment is used to characterize the dispersion of the color distribution, and the third-order moment is used to characterize the symmetry of the color distribution.

[0023] The chicken foot and comb image regions are optimized based on the low-order moments.

[0024] As an optional implementation, in the first aspect of the present invention, the breeding method further includes:

[0025] Acquire side images of breeder chickens in the third growth stage, and perform noise reduction and contrast enhancement on the side images of breeder chickens;

[0026] Key coordinate points of the keel region are identified using a pre-trained keel recognition model; wherein the keel recognition model is constructed using the HRNet model.

[0027] The keel centerline is fitted using a random sample consensus algorithm, and the average distance from each detection point to the centerline is calculated. The straightness of the keel is then determined based on the average distance.

[0028] As an optional implementation, in the first aspect of the present invention, the breeding method further includes:

[0029] Images of breeder chickens at different ages in the third growth stage are obtained, and a semantic segmentation model is used to extract the mask of the feather-covered area from the images of breeder chickens at different ages.

[0030] Compare feather masks of different ages and calculate the change in feather coverage per unit time.

[0031] The texture complexity of the feather surface is evaluated using texture analysis algorithms;

[0032] Identify holes or low-coverage areas in the feather mask, calculate the proportion of abnormal areas, and determine whether there are growth defects based on the proportion of abnormal areas. If there are no growth defects, determine the feather score based on the change in feather coverage, texture complexity, and proportion of abnormal areas.

[0033] The breeding method further includes:

[0034] Images of breeder chickens at different ages in the third growth stage are acquired. The body surface area and body length of the breeder chickens are calculated from the images of breeder chickens at different ages. A weight prediction model based on historical data is then established to determine the weight of the corresponding breeder chickens.

[0035] Based on the breed characteristics, a standard weight curve for each age is generated. The deviation between the actual weight and the standard value is compared. The deviation is used to determine whether the weight is within the normal fluctuation range, and the weight score is determined based on the deviation.

[0036] The breeding method further includes:

[0037] The corresponding bone score is determined based on the straightness of the keel, and the comprehensive health index of the third growth stage is obtained based on the bone score, feather score, weight score, and a pre-built health scoring model.

[0038] As an optional implementation, in the first aspect of the present invention, after determining the corresponding skeletal score based on the keel straightness, and obtaining the comprehensive health index of the third growth stage based on the skeletal score, feather score, weight score, and a pre-constructed health scoring model, the method further includes:

[0039] Breeding chickens are graded according to the comprehensive health index of the third growth stage to obtain breeding chicken grades, which include grade one, grade two, grade three and grade four.

[0040] If the breeder chicken is grade one, the corresponding breeder chicken will be kept for breeding. If the breeder chicken is grade four, the corresponding breeder chicken will be culled. Based on the above breeding results, breeder chickens of grade three and / or four will be kept for breeding or culled.

[0041] Assign numbers to the corresponding breeding chickens, including breed number, generation number, family number, hen number, and hatching egg number, and generate pedigrees by establishing families according to family ratios and excluding blood relations.

[0042] As an optional implementation, in the first aspect of the present invention, the breeding method further includes:

[0043] In the third growth stage, real-time motion images are acquired, and it is determined whether the corresponding hen has entered the core laying area. When the corresponding hen is detected to have entered the core laying area, the time the hen stays in the core laying area is counted, and the movement sequence data of the hen within a set time range before entering the core laying area is acquired.

[0044] The movement sequence data is input into the LSTM model to output the probability of the corresponding hen's pre-laying activity. When the probability of pre-laying activity is greater than the first set value and the time spent in the core laying area is greater than the second set value, multiple frames of core laying area images are continuously captured.

[0045] The key point recognition model is used to identify the key points of the hen in the core egg-laying area image to determine the key points of the hen, and the hen posture parameters are determined based on the key points. The hen posture parameters include the hip joint angle, knee joint angle, ankle joint angle and the angle between the tail and the body midline.

[0046] If the hen's posture parameters match the egg-laying behavior parameters, it is determined to be a valid egg-laying behavior, and the timestamp and cage ID are recorded, and the corresponding cage is updated.

[0047] A second aspect of this invention discloses a breeding system based on multi-dimensional data, comprising:

[0048] First acquisition module: used to acquire image information of chicks in the first growth stage, the chick image information including feather image information; wherein, the feather image information includes chick feather images and reference object images;

[0049] Feature recognition module: used to perform feature recognition on the feather image information to determine the reference object image in the feather image information, determine the pixel information of the corresponding reference object in the reference object image, and determine the pixel size based on the pixel information and the pre-stored reference object size;

[0050] Recognition module: used to preprocess the feather image information and input it into the feather feature recognition model to perform feature recognition to determine the corresponding feather feature information, and to determine the feather feature parameters of the chick based on the feather feature information and pixel size. The feather feature parameters include the length of the primary wing feathers and the length of the primary wing coverts.

[0051] Primary screening module: used to determine the sex of the chicks and predict growth parameters based on the length of the primary wing feathers and the length of the primary wing coverts, to perform primary screening based on the predicted growth parameters, and to carry out the next stage of breeding management based on the results of the primary screening.

[0052] The second acquisition module is used to acquire image information of breeder chickens in the second growth stage, input the image information of breeder chickens into the feature recognition model to acquire the feather image, foot image and comb image of the corresponding breeder chicken, and determine the feather color and foot color of the breeder chicken;

[0053] Secondary screening module: used to input the rooster comb image into the rooster comb recognition model to determine the corresponding rooster comb type; if the feather color, foot color and rooster comb type match the breed parameters of the corresponding breed chicken, then the chicken is retained for breeding and put into cages, and the corresponding individual characteristic parameters are recorded.

[0054] A third aspect of the present invention discloses an electronic device, comprising: a memory storing executable program code; a processor coupled to the memory; the processor calling the executable program code stored in the memory to execute the breeding method based on multi-dimensional data disclosed in the first aspect of the present invention.

[0055] A fourth aspect of the present invention discloses a computer-readable storage medium storing a computer program, wherein the computer program causes a computer to execute the breeding method based on multi-dimensional data disclosed in the first aspect of the present invention.

[0056] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0057] The solution in this embodiment of the invention constructs a precise breeding system for the entire cycle from chicks to breeder chickens by deeply integrating image recognition technology with multi-dimensional phenotypic data. It realizes automated sex identification, standardized breed verification, and intelligent growth prediction, providing an efficient management method for modern poultry breeding. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 This is a flowchart illustrating the breeding method based on multi-dimensional data disclosed in an embodiment of the present invention;

[0060] Figure 2 This is a schematic diagram of cage position monitoring disclosed in an embodiment of the present invention;

[0061] Figure 3 This is a schematic diagram illustrating the display of individual information as disclosed in an embodiment of the present invention;

[0062] Figure 4This is a schematic diagram showing the pedigree chart disclosed in the embodiments of the present invention;

[0063] Figure 5 This is a schematic diagram of a breeding system based on multi-dimensional data provided in an embodiment of the present invention;

[0064] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0065] 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 embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0066] It should be noted that the terms "first," "second," "third," "fourth," etc., in the specification and claims of this invention are used to distinguish different objects, not to describe a specific order. The terms "comprising" and "having," and any variations thereof, in the embodiments of this invention are intended to cover non-exclusive inclusion. Exemplarily, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or devices.

[0067] Example 1

[0068] Please see Figure 1 , Figure 1 This is a flowchart illustrating the breeding method based on multi-dimensional data disclosed in an embodiment of the present invention. The execution entity of the method described in this embodiment is an execution entity composed of software and / or hardware. This execution entity can receive relevant information via wired or / or wireless means and can send certain instructions. It may also have certain processing and storage functions. This execution entity can control multiple devices, such as remote physical servers or cloud servers and related software, or local hosts or servers and related software that perform related operations on devices located in a certain location. In some scenarios, it can also control multiple storage devices, which may be placed in the same location as the devices or in different locations. Figures 1 to 4 As shown, this breeding method based on multi-dimensional data includes the following steps:

[0069] S101: Acquire image information of chicks in the first growth stage, wherein the chick image information includes feather image information; wherein the feather image information includes chick feather images and reference object images;

[0070] S102: Perform feature recognition on the feather image information to determine a reference object image in the feather image information, determine the pixel information of the corresponding reference object in the reference object image, and determine the pixel size based on the pixel information and the pre-stored reference object size;

[0071] S103: After preprocessing the feather image information, it is input into the feather feature recognition model to perform feature recognition to determine the corresponding feather feature information, and the feather feature parameters of the chick are determined according to the feather feature information and the pixel size. The feather feature parameters include the length of the primary wing feathers and the length of the primary wing coverts.

[0072] S104: Determine the sex and predicted growth parameters of the chicks based on the primary wing feather length and the primary wing coverts length, perform primary screening based on the predicted growth parameters, and carry out the next stage of breeding management based on the primary screening results.

[0073] S105: Obtain image information of breeding chickens in the second growth stage, input the image information of breeding chickens into the feature recognition model to obtain the feather image, foot image and comb image of the corresponding breeding chicken, and determine the feather color and foot color of the breeding chicken;

[0074] S106: Input the rooster comb image into the rooster comb recognition model to determine the corresponding rooster comb type; if the feather color, foot color and rooster comb type match the breed parameters of the corresponding breed chicken, then carry out the breeding and cage-keeping operations, and record the corresponding individual characteristic parameters.

[0075] In this embodiment of the invention, specifically, the primary wing feather length is the average of the lengths of the third and fourth primary wing feathers and the overlying primary wing feathers counted from the wingtip, and the average feather length and the difference in feather length are calculated. By analyzing the proportional relationship between the primary wing feather length and the overlying primary wing feather length, the sex of chicks can be quickly and non-contactly identified. Compared with the traditional vent examination method, this method reduces stress and lowers labor costs, making it particularly suitable for large-scale automated farming scenarios.

[0076] This invention utilizes image recognition technology to quickly exclude hybrid or mutated individuals, ensuring the genetic purity of the breeding flock. Based on the matching degree between phenotypic characteristics such as feather color, foot color, and comb type and breed standard parameters, precise selection of breeding chickens for propagation is achieved. Automated feature recognition replaces manual visual judgment, reducing subjective errors and improving the accuracy and consistency of breeding decisions.

[0077] In this embodiment of the invention, a mapping relationship between pixels and actual size is established by using a reference image (such as a ruler of known size). Parameters such as the length of the main wing feathers can be calculated directly from the image, avoiding the time-consuming and error-prone problems of traditional manual measurement.

[0078] By integrating image data from multiple body parts, including feathers, feet, and combs, an individual characteristic map is constructed based on dimensions such as appearance and color features. Compared to single-dimensional breeding (such as focusing only on body weight), this method can capture breed-specific phenotypes. For example, foot color (such as yellow or blue) is an important characteristic of some breeds (such as Gushi chicken and Sanhuang chicken); comb type (such as single comb or bean comb) is associated with traits such as egg production performance and cold resistance, and can be used as an auxiliary indicator for breeding.

[0079] The preprocessed feather images are input into the feather feature recognition model, and combined with deep learning algorithms (such as convolutional neural networks) to automatically extract subtle features such as rachis texture and vane distribution, breaking through the limitations of human experience and realizing the quantitative analysis of latent traits (such as feather health).

[0080] Record individual characteristic parameters (such as wing length and comb type) and associate them with breeding stages to form a digital breeding profile. This facilitates tracing the bloodline of breeder chickens, analyzing the genetic patterns of traits, providing data support for optimizing subsequent breeding programs, and promoting the transformation from experience-based breeding to data-driven breeding.

[0081] More preferably, the step of determining the sex of the chick and predicting growth parameters based on the primary wing feather length and the primary wing coverts length includes:

[0082] S1021: Determine the average length and the length difference based on the length of the primary wing feather and the length of the overlying primary wing feather;

[0083] S1022: Determine the sex of the chick based on the length of the primary wing feathers and the length of the primary wing coverts; if the length difference and the average length meet the set conditions, the chick is determined to be a female chick; if the length difference and the average length do not meet the set conditions, the chick is determined to be a male chick.

[0084] S1023: Determine the first feather maturity of the corresponding female chick based on the average value of the primary wing feather length, the primary wing coverts length and length;

[0085] S1024: Determine the maturity of the second feathers of the corresponding male chick based on the average length and length of the primary wing coverts.

[0086] In this embodiment of the invention, the discrimination criteria are constructed by combining the length difference (primary wing feather length - primary wing covert length) and the mean length, rather than relying on a single indicator (such as solely on the length difference), which can more comprehensively reflect the variety-specific growth patterns. For example:

[0087] For certain fast-feathering breeds (where the primary wing feathers of female chicks are longer than the coverts), setting a length difference of >5mm and a mean of >12mm as the condition for female chicks can eliminate misjudgments due to insufficient difference caused by individual developmental delays. For male chicks (slow-feathering breeds), filtering out abnormally small individuals by using the mean can avoid misjudgments of sex due to physiological defects. Cross-validation using two indicators can significantly improve the accuracy of sex identification compared to the traditional single-morphological observation method.

[0088] The first feather maturity is calculated using the primary wing feather length, primary wing coverts length, and their mean values ​​(e.g., maturity = (primary wing feather length + primary wing coverts length) / (2 × mean)), reflecting the balance of wing development. A maturity value closer to 1 indicates better feather growth synchronization, predicting stronger thermoregulation and higher disease resistance in later stages. The second feather maturity is calculated based on the primary wing coverts length and mean values ​​(e.g., maturity = primary wing coverts length / mean), focusing on assessing the growth rate of the coverts. A higher maturity value indicates faster feather coverage efficiency, reducing energy loss and promoting weight gain. Feather maturity is significantly positively correlated with growth parameters (e.g., 7-day-old weight and 42-day-old slaughter weight).

[0089] A chick growth potential scoring system can be constructed using maturity indicators (e.g., dividing maturity into four levels: S / A / B / C). Combined with historical breeding data, a prediction model can be trained to predict the late-stage growth performance of individual chicks (e.g., peak egg production and slaughter rate) 6-8 weeks in advance, providing a basis for refined breeding (e.g., group feeding and focused breeding).

[0090] In traditional breeding, the culling rate due to similar appearance is about 8%-10%. By introducing a maturity quantification index, outliers can be filtered out through data thresholds, thus reducing the culling rate. For example, in a batch of chicks, individuals with length differences close to the critical value can be identified by maturity indicators as either developmentally delayed but of the correct sex or a misjudgment of sex, preventing healthy female chicks from being mistakenly culled.

[0091] More preferably, the step of inputting the breeder chicken image information into the feature recognition model to obtain the corresponding breeder chicken's feather image, foot image, and comb image, and determining the breeder chicken's feather color and foot color, includes:

[0092] The image information of the breeding chickens is input into the backbone network to extract convolutional feature maps, and a series of anchor frames of preset size and proportion are generated on the convolutional feature maps using a sliding window;

[0093] For each anchor box, predict the probability that it contains feathers, foot, or comb foreground, filter out low-probability regions, and fine-tune the coordinates of the foreground anchor boxes to obtain the corresponding chicken feather, foot, and comb images.

[0094] The feather image is processed using EfficientNetV2-L as the backbone network, and shallow feather features and deep feather color features are extracted to optimize the feather image region.

[0095] Convert the obtained chicken foot and comb images from RGB to HSV or Lab color space;

[0096] For each color channel in the converted color space, the lower-order moments of the corresponding image are calculated. The lower-order moments include the first-order moment, the second-order moment, and the third-order moment. The first-order moment is used to characterize the average value of the color channel, the second-order moment is used to characterize the dispersion of the color distribution, and the third-order moment is used to characterize the symmetry of the color distribution.

[0097] The chicken foot and comb image regions are optimized based on the low-order moments.

[0098] This invention generates anchor frames of various sizes and scales (such as elongated anchor frames for feather areas and circular anchor frames for comb areas) through a sliding window, adapting to the morphological differences of different breeds of chickens. Foreground probability filtering and coordinate fine-tuning mechanisms can eliminate invalid anchor frames, reducing positioning error accuracy. An EfficientNetV2-L backbone network replaces the traditional CNN, and through hybrid depthwise convolution and dynamic feature scaling, it can simultaneously capture shallow texture features of feathers (such as shaft thickness) and deep color features (such as feather tip pigment distribution). The EfficientNetV2 series optimizes network width, depth, and resolution through a compound scaling strategy, and its lightweight design and multi-stage feature output characteristics are suitable for multi-scale feature extraction.

[0099] Specifically, shallow networks (early convolutional layers) offer high resolution and a small receptive field, excelling at capturing local details (such as feather edge texture and fine structures). Shallow feature extraction (feather edge details) and input preprocessing: High-resolution input (e.g., 384×384 or higher) is used for the feather image to prevent premature loss of edge details due to downsampling. Shallow convolutional layer design: Initial convolutional layers (e.g., Stem modules): 3×3 convolutions or depthwise separable convolutions are used to preserve edge information with low computational cost. Premature pooling is avoided: Max pooling or strided convolutions (StridedConv) are reduced in early layers; feature map resolution is maintained through dilated convolutions (DilatedConv) or convolutions with a stride of 1. Edge detection branch is introduced: An edge detection subnetwork (e.g., a Canny-based or deep learning-based edge detector) is added in parallel to the shallow network, sharing low-level features with the main network and directly outputting a feather edge probability map. Feature enhancement module: An attention mechanism (such as SpatialAttention) is added after shallow convolution to enhance the feature response of edge regions through weights and suppress background noise.

[0100] Deep networks (later convolutional layers): low resolution, large receptive field, adept at extracting global semantics (such as overall feather color distribution, feather shape contours). Deep networks expand their receptive field through hierarchical downsampling: in the later stages of the network, the feature map resolution is gradually reduced (e.g., from 384×384 to 12×12) using strided convolutional or pooling layers, expanding the receptive field to capture global feather color distribution.

[0101] Feature Pyramid Structure: Utilizing structures such as FPN (Feature Pyramid Network) or BiFPN, a multi-scale feature pyramid is constructed in the deep network to fuse global semantic information from different levels (such as the dominant hue and distribution pattern of feather color). Input Color Space Adjustment: The RGB image is converted to HSV or Lab color space to separate luminance and chromaticity information, allowing the deep network to focus on extracting color features.

[0102] Non-local operations or Transformer modules: Inserting non-local convolutions or VisionTransformer modules into deep networks models long-distance dependencies (such as spatial correlations of feather color in different regions) and improves the representation ability of global features.

[0103] Cross-layer feature fusion and skip connections directly add or concatenate high-resolution edge features in shallow layers with low-resolution semantic features in deep layers through skip connections.

[0104] Feature Pyramid Fusion (FPFusion): By upsampling deep feature maps to shallow resolution, adjusting the number of channels using 1×1 convolutions, and then element-wise adding or concatenating with shallow features, a mixed feature map containing both detailed and global information is generated. Edge detection loss (e.g., BCELoss) and feather color classification loss (e.g., Cross-EntropyLoss) are designed to supervise the learning of shallow and deep features respectively, ensuring the effectiveness of both types of features. Progressive training: First, the shallow network is trained to converge its edge detail extraction ability. Then, the shallow weights are frozen, and the deep network is trained to learn global features. Finally, the entire network is jointly optimized to avoid gradient vanishing or overfitting.

[0105] In this embodiment of the invention, the color moment feature directly characterizes the color distribution, making it suitable for distinguishing areas with significant color differences (such as red combs and mottled feathers). Combining the two can more comprehensively characterize the target features, improving the detection accuracy of the RPN network for combs and feathers, and is especially suitable for scenarios such as early sex identification and feather development assessment in chicks.

[0106] Color moment features capture the global distribution characteristics of colors by calculating statistics such as the mean, variance, and third moment (skewness) of the image's color channels. In rooster comb and feather detection, color moment features can be used to: distinguish between rooster combs and feathers: rooster combs are usually red (due to abundant capillaries), while feathers are diverse in color (such as mottled and white), and color moments can directly describe this difference. Identify feather color types: for example, the color distribution of feather markings in Qingyuan Ma chickens can be quantified using color moments.

[0107] Color space conversion: Converting an image from RGB to color spaces such as HSV and Lab that are more in line with human visual perception (e.g., using the H channel to represent hue).

[0108] Calculate the moments of each order: For each color channel (e.g., H, S, V):

[0109] First moment (mean): Represents the average hue of a color (such as the mean red of a rooster's comb).

[0110] Second moment (variance): Represents the degree of dispersion of color (such as the uniformity of color in feather markings).

[0111] Third moment (skewness): Represents the symmetry of color distribution (e.g., whether it is reddish or yellowish).

[0112] Higher-order moments (optional): Further describe the details of the color distribution.

[0113] Generate color moment feature vectors: concatenate the moment values ​​of each channel (such as the mean, variance, and third moment of the H, S, and V channels, totaling 9 dimensions) to characterize the color distribution.

[0114] Cockscomb: The red channel in the color moment feature has a high mean and a small variance (the color is relatively uniform).

[0115] Feathers: Color moment characteristics may exhibit multi-channel mixing (e.g., the H channel mean of mottled feathers has a wide distribution and a large variance).

[0116] In practical applications, HOG features and color moment features are often fused together to combine the complementarity of shape, texture, and color information: HOG features capture the edge and texture structure of a target by calculating the gradient orientation histogram of local image regions, and are sensitive to the shape and contour of the target. In rooster comb and feather detection, HOG features can be used to: distinguish the rooster comb from surrounding tissue: rooster combs typically have distinct edges (such as serrated outlines) and textures (such as wrinkles), and HOG can capture these gradient changes. Identify the direction and distribution of feathers: feathers have directionality (such as the orientation of the rachis) and regular texture, and HOG can describe their structure through the distribution of gradient orientations.

[0117] More preferably, the breeding method further includes:

[0118] Acquire side images of breeder chickens in the third growth stage, and perform noise reduction and contrast enhancement on the side images of breeder chickens;

[0119] Key coordinate points of the keel region are identified using a pre-trained keel recognition model; wherein the keel recognition model is constructed using the HRNet model.

[0120] The keel centerline is fitted using a random sample consensus algorithm, and the average distance from each detection point to the centerline is calculated. The straightness of the keel is then determined based on the average distance.

[0121] This invention, through a parallel multi-resolution branching structure, preserves fine spatial information of the image and improves the positioning accuracy of key points in the keel region (such as the keel tip, sternal prominence, and keel tip). Noise reduction and contrast enhancement (such as adaptive histogram equalization CLAHE) significantly improve the clarity of the keel outline under feather cover.

[0122] The Random Sample Consensus (RANSAC) algorithm iteratively removes outliers (such as feather interference points and image noise points) to fit the optimal keel centerline. This technology, through HRNet high-precision keypoint detection and the robust RANSAC fitting algorithm, achieves a quantitative assessment of the straightness of the keel in breeder chickens, providing objective and traceable morphological indicators for breeding decisions. Its core value lies in early identification of high-quality breeding stock, reducing breeding risks, and accelerating the genetic improvement process, making it particularly suitable for the precision management of modern breeder farms. Combined with early sex identification and feather color analysis, it forms a full-cycle intelligent breeding solution from chicks to adults. The RANSAC algorithm is used to fit the keel centerline, calculating the average distance from each detection point to the centerline; a smaller distance indicates a straighter keel. The average distance is converted to a score of 0-100 (e.g., deducting 5 points for every 1 mm increase in distance); a higher score indicates better skeletal development.

[0123] More preferably, the breeding method further includes:

[0124] Images of breeder chickens at different ages in the third growth stage are obtained, and a semantic segmentation model is used to extract the mask of the feather-covered area from the images of breeder chickens at different ages.

[0125] Compare feather masks of different ages and calculate the change in feather coverage per unit time.

[0126] The texture complexity of the feather surface is evaluated using texture analysis algorithms;

[0127] Identify holes or low-coverage areas in the feather mask, calculate the proportion of abnormal areas, and determine whether there are growth defects based on the proportion of abnormal areas. If there are no growth defects, determine the feather score based on the change in feather coverage, texture complexity, and proportion of abnormal areas.

[0128] The breeding method further includes:

[0129] Images of breeder chickens at different ages in the third growth stage are acquired. The body surface area and body length of the breeder chickens are calculated from the images of breeder chickens at different ages. A weight prediction model based on historical data is then established to determine the weight of the corresponding breeder chickens.

[0130] Based on the breed characteristics, a standard weight curve for each age is generated. The deviation between the actual weight and the standard value is compared. The deviation is used to determine whether the weight is within the normal fluctuation range, and the weight score is determined based on the deviation.

[0131] The breeding method further includes:

[0132] The corresponding bone score is determined based on the straightness of the keel, and the comprehensive health index of the third growth stage is obtained based on the bone score, feather score, weight score, and a pre-built health scoring model.

[0133] Images of breeder chickens at different ages in the third growth stage are acquired. The body surface area (by segmentation mask) and body length (by keypoint distance) of the breeder chickens are calculated from the images of breeder chickens at different ages. A weight prediction model (such as linear regression or machine learning model) is established by combining historical data.

[0134] Standard curve matching: Generate a standard weight curve for each age group based on breed characteristics (e.g., the standard weight for broiler breeder chickens at 20 weeks of age is 3kg), and compare the deviation of the actual weight from the standard value.

[0135] Statistical analysis: Calculate the Z-score (multiple of the standard deviation from the group mean) or percentage deviation of body weight to determine whether an individual's weight is within the normal fluctuation range (e.g., ±10% is normal). Assign weights to bones (40%), feathers (30%), and body weight (30%), and calculate a comprehensive health index (e.g., bone score × 0.4 + feather score × 0.3 + body weight score × 0.3).

[0136] This invention acquires images of breeder chickens and uses a semantic segmentation model to extract feather coverage area masks, accurately defining feather regions. By comparing feather masks at different ages to calculate changes in feather coverage, the growth rate of feathers can be intuitively understood; assessing texture complexity reflects feather quality; identifying holes or low-coverage areas and calculating the proportion of abnormal areas can accurately determine whether there are defects in feather growth. Combining these indicators to determine a feather score provides a comprehensive and accurate assessment of the feather growth status of breeder chickens.

[0137] Feather growth is an important indicator of the health and quality of breeder chickens. Accurate assessment of feather condition helps to select breeder chickens with good feather growth and excellent health, avoiding the impact of feather problems on the reproductive performance of breeder chickens or the quality of offspring, thus improving the accuracy and scientific nature of breeder selection.

[0138] This non-contact weight measurement method calculates body surface area and length from breeder chicken images and uses a weight prediction model built from historical data to determine breeder chicken weight. This avoids the stress caused by manual weighing and allows for the rapid and accurate acquisition of a large amount of breeder chicken weight data. A weight standard curve is generated based on breed characteristics, and comparing the actual weight with the standard value provides a clear understanding of the breeder chicken's weight development, determines whether it falls within the normal fluctuation range, and assigns a weight score. This helps to promptly identify breeder chickens with abnormal weight, allowing for targeted adjustments to feeding and management measures, ensuring the consistency and uniformity of the breeder chicken population's weight, and improving overall production performance.

[0139] By substituting bone score, feather score, and weight score into a pre-constructed health scoring model, a comprehensive health index for the third growth stage is obtained. This index comprehensively considers multiple important physiological indicators of breeder chickens, fully reflecting their health status. It provides a comprehensive and quantitative evaluation standard for breeder chicken selection, helping to select breeder chickens with the best health, improve the reproductive performance and offspring quality of breeder chickens, and thus enhance the overall production performance and economic benefits of the flock.

[0140] More preferably, after determining the corresponding skeletal score based on the keel straightness, and obtaining the comprehensive health index of the third growth stage based on the skeletal score, feather score, weight score, and a pre-built health scoring model, the method further includes:

[0141] Breeding chickens are graded according to the comprehensive health index of the third growth stage to obtain breeding chicken grades, which include grade one, grade two, grade three and grade four.

[0142] If the breeder chicken is grade one, the corresponding breeder chicken will be kept for breeding. If the breeder chicken is grade four, the corresponding breeder chicken will be culled. Based on the above breeding results, breeder chickens of grade three and / or four will be kept for breeding or culled.

[0143] Assign numbers to the corresponding breeding chickens, including breed number, generation number, family number, hen number, and hatching egg number, and generate pedigrees by establishing families according to family ratios and excluding blood relations.

[0144] This invention classifies breeder chickens into four grades based on a comprehensive health index. Grade 1 breeder chickens are directly retained for breeding to fully utilize their superior genes; Grade 4 breeder chickens are culled to avoid wasting resources on individuals of poor quality. For Grade 3 breeder chickens, flexible decisions are made regarding retention or culling, achieving efficient utilization of breeder chicken resources, ensuring the overall quality of the breeder flock is improved, optimizing the genetic structure of the flock, and laying a solid foundation for subsequent production. This multi-level design approach allows for more flexible selection of breeder chickens in later stages.

[0145] Clear grading and corresponding retention / culling rules make the breeding process of breeder chickens more standardized and regulated. This avoids the arbitrariness and uncertainty of subjective human judgment, enabling the rapid and accurate selection of qualified breeder chickens, shortening the breeding cycle, improving breeding efficiency, and ensuring the accuracy and reliability of the breeding results, thus contributing to the development of higher-quality breeder chickens.

[0146] More preferably, the breeding method further includes:

[0147] In the third growth stage, real-time motion images are acquired, and it is determined whether the corresponding hen has entered the core laying area. When the corresponding hen is detected to have entered the core laying area, the time the hen stays in the core laying area is counted, and the movement sequence data of the hen within a set time range before entering the core laying area is acquired.

[0148] The movement sequence data is input into the LSTM model to output the probability of the corresponding hen's pre-laying activity. When the probability of pre-laying activity is greater than the first set value and the time spent in the core laying area is greater than the second set value, multiple frames of core laying area images are continuously captured.

[0149] The key point recognition model is used to identify the key points of the hen in the core egg-laying area image to determine the key points of the hen, and the hen posture parameters are determined based on the key points. The hen posture parameters include the hip joint angle, knee joint angle, ankle joint angle and the angle between the tail and the body midline.

[0150] If the hen's posture parameters match the egg-laying behavior parameters, it is determined to be a valid egg-laying behavior, and the timestamp and cage ID are recorded, and the corresponding cage is updated.

[0151] The solution in this invention acquires motion images in real time and analyzes the hen's movement sequence data (such as walking path and speed changes before entering the core laying area), then uses an LSTM model to predict the probability of pre-laying activity. The LSTM model's ability to model long-term dependencies in time-series data effectively captures pre-laying behavioral patterns in hens (such as frequent pacing and pecking), avoiding the lag and subjectivity of traditional manual observation. Combined with a threshold for the time spent in the core laying area (greater than a second set value), invalid dwelling behaviors (such as accidental entry into resting areas) are filtered out, ensuring that subsequent image acquisition is triggered only for high-probability laying behaviors, thus improving data processing efficiency.

[0152] Keypoint recognition models are used to extract the hen's joints (such as the hip and knee joints) and calculate posture parameters (such as joint angles and tail angles). These parameters are highly correlated with the biological characteristics of egg-laying behavior (such as crouching and tail raising), accurately distinguishing egg-laying behavior from other activities (such as preening and feeding). Only when the posture parameters match preset egg-laying behavior parameters is the behavior considered valid, avoiding misjudgments. For example, an angle between the tail and the body's midline exceeding a specific threshold (such as 45°) is often accompanied by cloaca exposure, a key characteristic of egg-laying.

[0153] Once a valid egg-laying behavior is triggered, the system automatically records the timestamp (accurate to the second), cage ID, and updates the cage status (e.g., marked as "laying eggs"). This data provides real-time and accurate basic data for analyzing the reproductive performance of breeder hens (e.g., egg-laying frequency per cage, peak egg-laying period), supporting refined feeding management (e.g., adjusting lighting time and feed nutrition).

[0154] In practice, pressure sensors can be installed at the bottom of the cage to help verify egg-laying events (such as the impact signal when an egg rolls down) through pressure surges.

[0155] Existing egg counting methods require daily collection and statistical analysis, which is prone to omissions and inefficient. The solution in this application directly associates the corresponding behavior with the corresponding cage location, enabling direct data counting and updates, which is quick, convenient, and highly accurate. This invention primarily targets Ma Huang chickens.

[0156] The scheme of this invention embodiment can also perform a fourth selection, the selection parameters of which can be the body weight at 300 days of age, the egg production of hens from the start of laying to 300 days of age, the weight of eggs at 300 days of age, the weight of eggs at 301 days of age, and the weight of eggs at 302 days of age; and the selection is carried out by comprehensively using the above information.

[0157] In addition to the methods mentioned above, the eggs of breeding chickens after mating can also be analyzed, mainly focusing on the dynamic viscoelasticity of the albumen. The eggs are placed on a rotating instrument and then illuminated. The reflux rate of the concentrated albumen is tracked by taking images, mainly through the inter-frame gradient change. The dynamic viscoelasticity analysis of the albumen achieves the lossless quantization of Hough units through the rotating illumination time-series imaging technology.

[0158] Specifically, the concentrated albumen of a fresh egg has a gel-like structure. After the external force is removed, it will slowly flow back (viscoelastic recovery). Since the light transmittance of the thin albumen is high, the image is bright. The light transmittance of the concentrated albumen is low, resulting in dark areas in the image. Rotation triggers rheology: when the egg rotates, the centrifugal force deforms the albumen. After it stops, the concentrated albumen flows back to the geometric center. Due to the existence of the above mechanism, the stress relaxation time constant of the albumen can be measured.

[0159] Specifically, the viscoelastic parameters are calculated as follows: Where: A0 is the peak displacement of the centroid at the instant the rotation stops, A t It is the displacement of the center of mass at time t, when A t When / A0 = 1 / e, the corresponding time τ is the characteristic recovery time; the strength of protein cross-linking can be characterized by the corresponding τ; thus, the efficiency of embryonic protein absorption can be determined, and the actual developmental status of the hatching egg can be evaluated. In specific implementation, data conversion between viscoelastic parameters and Hough units can also be achieved.

[0160] The solution in this embodiment of the invention constructs a precise breeding system for the entire cycle from chicks to breeder chickens by deeply integrating image recognition technology with multi-dimensional phenotypic data. It realizes automated sex identification, standardized breed verification, and intelligent growth prediction, providing an efficient management method for modern poultry breeding.

[0161] Example 2

[0162] Please see Figure 5 , Figure 5 This is a schematic diagram of the breeding system based on multi-dimensional data disclosed in an embodiment of the present invention. Figure 5 As shown, this breeding system based on multi-dimensional data may include:

[0163] First acquisition module 21: used to acquire image information of chicks in the first growth stage, the chick image information including feather image information; wherein, the feather image information includes chick feather images and reference object images;

[0164] Feature recognition module 22: used to perform feature recognition on the feather image information to determine the reference object image in the feather image information, determine the pixel information of the corresponding reference object in the reference object image, and determine the pixel size based on the pixel information and the pre-stored reference object size;

[0165] Recognition module 23: is used to preprocess the feather image information and input it into the feather feature recognition model to perform feature recognition to determine the corresponding feather feature information, and to determine the feather feature parameters of the chick based on the feather feature information and pixel size. The feather feature parameters include the length of the primary wing feathers and the length of the primary wing coverts.

[0166] Primary screening module 24: used to determine the sex of the corresponding chick and predict growth parameters based on the length of the primary wing feathers and the length of the primary wing coverts, to perform primary screening based on the predicted growth parameters, and to carry out the next stage of breeding management based on the results of the primary screening.

[0167] The second acquisition module 25 is used to acquire image information of breeding chickens in the second growth stage, input the image information of breeding chickens into the feature recognition model to acquire the feather image, foot image and comb image of the corresponding breeding chicken, and determine the feather color and foot color of the breeding chicken.

[0168] Secondary screening module 26: used to input the rooster comb image into the rooster comb recognition model to determine the corresponding rooster comb type; if the feather color, tarsus color and rooster comb type match the breed parameters of the corresponding breed chicken, then the chicken is retained for breeding and put into cages, and the corresponding individual characteristic parameters are recorded.

[0169] The solution in this embodiment of the invention constructs a precise breeding system for the entire cycle from chicks to breeder chickens by deeply integrating image recognition technology with multi-dimensional phenotypic data. It realizes automated sex identification, standardized breed verification, and intelligent growth prediction, providing an efficient management method for modern poultry breeding.

[0170] Example 3

[0171] Please see Figure 6 , Figure 6 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of the present invention. The electronic device can be a computer, a server, etc. Of course, in certain cases, it can also be a mobile phone, tablet computer, monitoring terminal, or other smart device, as well as an image acquisition device with processing capabilities. Figure 6 As shown, the electronic device may include:

[0172] Memory 510 storing executable program code;

[0173] Processor 520 coupled to memory 510;

[0174] The processor 520 calls the executable program code stored in the memory 510 to execute some or all of the steps in the breeding method based on multi-dimensional data in Embodiment 1.

[0175] This invention discloses a computer-readable storage medium storing a computer program that causes a computer to perform some or all of the steps in the breeding method based on multi-dimensional data in Embodiment 1.

[0176] This invention also discloses a computer program product, wherein when the computer program product is run on a computer, the computer performs some or all of the steps in the breeding method based on multi-dimensional data in Embodiment 1.

[0177] This invention also discloses an application publishing platform, which is used to publish computer program products. When the computer program products are run on a computer, the computer performs some or all of the steps in the breeding method based on multi-dimensional data in Embodiment 1.

[0178] In various embodiments of the present invention, it should be understood that the sequence number of each process does not necessarily imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0179] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they can be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0180] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0181] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-accessible memory. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several requests to cause a computer device (which can be a personal computer, server, or network device, specifically a processor in the computer device) to execute some or all of the steps of the methods described in the various embodiments of the present invention.

[0182] In the embodiments provided by this invention, it should be understood that "B corresponding to A" means that B is associated with A, and B can be determined based on A. However, it should also be understood that determining B based on A does not mean determining B solely based on A; B can also be determined based on A and / or other information.

[0183] Those skilled in the art will understand that some or all of the steps in the various methods of the embodiments described can be implemented by a program instructing related hardware. This program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0184] The above provides a detailed description of the breeding method, system, electronic device, and storage medium based on multi-dimensional data disclosed in the embodiments of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A breeding method based on multi-dimensional data, characterized in that, include: Acquire image information of chicks in the first growth stage, wherein the chick image information includes feather image information; wherein the feather image information includes chick feather images and reference object images; Feature recognition is performed on the feather image information to determine a reference object image in the feather image information, the pixel information of the corresponding reference object in the reference object image is determined, and the pixel size is determined based on the pixel information and the pre-stored reference object size; After preprocessing the feather image information, it is input into the feather feature recognition model to identify the corresponding feather feature information. Based on the feather feature information and pixel size, the feather feature parameters of the chick are determined. The feather feature parameters include the length of the primary wing feathers and the length of the primary wing coverts. The sex of the chicks and the predicted growth parameters are determined based on the length of the primary wing feathers and the length of the primary wing coverts. Primary screening is then performed based on the predicted growth parameters, and the next stage of breeding management is carried out based on the results of the primary screening. Acquire image information of breeder chickens in the second growth stage, input the breeder chicken image information into the feature recognition model to obtain the corresponding breeder chicken feather image, foot image and comb image, and determine the feather color and foot color of the breeder chicken; The rooster comb image is input into the rooster comb recognition model to determine the corresponding rooster comb type; if the feather color, foot color and rooster comb type match the breed parameters of the corresponding breed chicken, then the chicken is kept for breeding and put into a cage, and the corresponding individual characteristic parameters are recorded.

2. The breeding method based on multi-dimensional data as described in claim 1, characterized in that, The process of determining the sex of the chick and predicting growth parameters based on the primary wing feather length and the primary wing coverts length includes: The mean length and the length difference are determined based on the lengths of the primary wing feathers and the overlying primary wing feathers. The sex of the chick is determined based on the length of the primary wing feathers and the length of the primary wing coverts; if the length difference and the average length meet the set conditions, the chick is determined to be a female chick; if the length difference and the average length do not meet the set conditions, the chick is determined to be a male chick. The first feather maturity of the corresponding female chick is determined based on the average length of the primary wing feathers, the length of the primary wing coverts, and the length of the primary wing coverts. The maturity of the second feathers of the corresponding male chicks is determined by the average length and length of the primary wing coverts.

3. The breeding method based on multi-dimensional data as described in claim 1, characterized in that, The step of inputting the breeder chicken image information into a feature recognition model to obtain the corresponding breeder chicken's feather image, foot image, and comb image, and determining the breeder chicken's feather color and foot color, includes: The image information of the breeding chickens is input into the backbone network to extract convolutional feature maps, and a series of anchor frames of preset size and proportion are generated on the convolutional feature maps using a sliding window; For each anchor box, predict the probability that it contains feathers, foot, or comb foreground, filter out low-probability regions, and fine-tune the coordinates of the foreground anchor boxes to obtain the corresponding chicken feather, foot, and comb images. The feather image is processed using EfficientNetV2-L as the backbone network, and shallow feather features and deep feather color features are extracted to optimize the feather image region. Convert the obtained chicken foot and comb images from RGB to HSV or Lab color space; For each color channel in the converted color space, the lower-order moments of the corresponding image are calculated. The lower-order moments include the first-order moment, the second-order moment, and the third-order moment. The first-order moment is used to characterize the average value of the color channel, the second-order moment is used to characterize the dispersion of the color distribution, and the third-order moment is used to characterize the symmetry of the color distribution. The chicken foot and comb image regions are optimized based on the low-order moments.

4. The breeding method based on multi-dimensional data as described in claim 1, characterized in that, The breeding method further includes: Acquire side images of breeder chickens in the third growth stage, and perform noise reduction and contrast enhancement on the side images of breeder chickens; Key coordinate points of the keel region are identified using a pre-trained keel recognition model; wherein the keel recognition model is constructed using the HRNet model. The keel centerline is fitted using a random sample consensus algorithm, and the average distance from each detection point to the centerline is calculated. The straightness of the keel is then determined based on the average distance.

5. The breeding method based on multi-dimensional data as described in claim 4, characterized in that, The breeding method further includes: Images of breeder chickens at different ages in the third growth stage are obtained, and a semantic segmentation model is used to extract the mask of the feather-covered area from the images of breeder chickens at different ages. Compare feather masks of different ages and calculate the change in feather coverage per unit time. The texture complexity of the feather surface is evaluated using texture analysis algorithms; Identify holes or low-coverage areas in the feather mask, calculate the proportion of abnormal areas, and determine whether there are growth defects based on the proportion of abnormal areas. If there are no growth defects, determine the feather score based on the change in feather coverage, texture complexity, and proportion of abnormal areas. The breeding method further includes: Images of breeder chickens at different ages in the third growth stage are acquired. The body surface area and body length of the breeder chickens are calculated from the images of breeder chickens at different ages. A weight prediction model based on historical data is then established to determine the weight of the corresponding breeder chickens. Based on the breed characteristics, a standard weight curve for each age is generated. The deviation between the actual weight and the standard value is compared. The deviation is used to determine whether the weight is within the normal fluctuation range, and the weight score is determined based on the deviation. The breeding method further includes: The corresponding bone score is determined based on the straightness of the keel, and the comprehensive health index of the third growth stage is obtained based on the bone score, feather score, weight score, and a pre-built health scoring model.

6. The breeding method based on multi-dimensional data as described in claim 5, characterized in that, After determining the corresponding skeletal score based on the keel straightness, and obtaining the comprehensive health index for the third growth stage based on the skeletal score, feather score, weight score, and a pre-built health scoring model, the process further includes: Breeding chickens are graded according to the comprehensive health index of the third growth stage to obtain breeding chicken grades, which include grade one, grade two, grade three and grade four. If the breeder chicken is grade one, the corresponding breeder chicken will be kept for breeding. If the breeder chicken is grade four, the corresponding breeder chicken will be culled. Breeder chickens of grade two and / or three will be kept for breeding or culled based on the breeding results. Assign numbers to the corresponding breeding chickens, including breed number, generation number, family number, hen number, and hatching egg number, and generate pedigrees by establishing families according to family ratios and excluding blood relations.

7. The breeding method based on multi-dimensional data as described in claim 1, characterized in that, The breeding method further includes: In the third growth stage, real-time motion images are acquired, and it is determined whether the corresponding hen has entered the core laying area. When the corresponding hen is detected to have entered the core laying area, the time the hen stays in the core laying area is counted, and the movement sequence data of the hen within a set time range before entering the core laying area is acquired. The movement sequence data is input into the LSTM model to output the probability of the corresponding hen's pre-laying activity. When the probability of pre-laying activity is greater than the first set value and the time spent in the core laying area is greater than the second set value, multiple frames of core laying area images are continuously captured. The key point recognition model is used to identify the key points of the hen in the core egg-laying area image to determine the key points of the hen, and the hen posture parameters are determined based on the key points. The hen posture parameters include the hip joint angle, knee joint angle, ankle joint angle and the angle between the tail and the body midline. If the hen's posture parameters match the egg-laying behavior parameters, it is determined to be a valid egg-laying behavior, and the timestamp and cage ID are recorded, and the corresponding cage is updated.

8. A breeding system based on multi-dimensional data, characterized in that, include: First acquisition module: used to acquire image information of chicks in the first growth stage, the chick image information including feather image information; wherein, the feather image information includes chick feather images and reference object images; Feature recognition module: used to perform feature recognition on the feather image information to determine the reference object image in the feather image information, determine the pixel information of the corresponding reference object in the reference object image, and determine the pixel size based on the pixel information and the pre-stored reference object size; Recognition module: used to preprocess the feather image information and input it into the feather feature recognition model to perform feature recognition to determine the corresponding feather feature information, and to determine the feather feature parameters of the chick based on the feather feature information and pixel size. The feather feature parameters include the length of the primary wing feathers and the length of the primary wing coverts. Primary screening module: used to determine the sex of the chicks and predict growth parameters based on the length of the primary wing feathers and the length of the primary wing coverts, to perform primary screening based on the predicted growth parameters, and to carry out the next stage of breeding management based on the results of the primary screening. The second acquisition module is used to acquire image information of breeder chickens in the second growth stage, input the image information of breeder chickens into the feature recognition model to acquire the feather image, foot image and comb image of the corresponding breeder chicken, and determine the feather color and foot color of the breeder chicken; Secondary screening module: used to input the rooster comb image into the rooster comb recognition model to determine the corresponding rooster comb type; if the feather color, foot color and rooster comb type match the breed parameters of the corresponding breed chicken, then the chicken is retained for breeding and put into cages, and the corresponding individual characteristic parameters are recorded.

9. An electronic device, characterized in that, include: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the breeding method based on multi-dimensional data as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein the computer program causes a computer to perform the breeding method based on multidimensional data as described in any one of claims 1 to 7.

Images