Poultry growth intelligent monitoring method and system based on multi-modal data fusion

Abstract

The invention provides an intelligent poultry growth monitoring method and system based on multi-modal data fusion, relates to the technical field of intelligent breeding, and solves the technical problems of incomplete information dimension, large sample deviation, poor dynamic adaptability and limited evaluation dimension caused by dependence on single sensor data in the existing poultry weight estimation technology. The method comprises the following steps: collecting multi-modal data of poultry through multiple sensors; de-noising, de-duplication and identification are carried out on the multi-modal data through a preprocessing algorithm to obtain standard data; establishing a multi-dimensional feature vector based on the standard data, and performing weighted fusion on the multi-dimensional feature vector through an attention mechanism to obtain a fusion feature; based on the fusion features, a GMM-HMM model and a Transform network are adopted to carry out poultry group growth dynamic modeling and trend prediction, a group uniformity evaluation index is calculated, and an analysis result is obtained; and generating a body weight estimation value and a growth early warning signal according to an analysis result. The method is used in the intelligent poultry breeding process.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent breeding, and in particular to a poultry growth intelligent monitoring method and system based on multi-modal data fusion. BACKGROUND

[0002] Traditional poultry group growth evaluation methods mainly rely on individual sampling weighing or automatic equipment based on a single weight sensor. Such methods have several inherent technical defects: the data source is single, only body weight information can be obtained, and key biological characteristics such as individual body shape structure and behavior activity cannot be fully reflected, resulting in a serious lack of evaluation dimensions; at the same time, the active individual data is over-sampled and the weak individual data is missing due to the active sampling mechanism, causing significant sample bias and making it difficult to truly represent the overall group condition; in addition, the existing algorithm parameters are fixed and lack the ability to dynamically adapt to group growth stages and environmental changes, with poor adaptability; finally, the evaluation results are usually only a single indicator such as average body weight, which cannot provide comprehensive and predictive insights into group uniformity, growth trend and health status, seriously restricting the implementation of precise management decisions. SUMMARY

[0003] The present application provides a poultry growth intelligent monitoring method and system based on multi-modal data fusion, which solves the technical problems of incomplete information dimension, large sample bias, poor dynamic adaptability and limited evaluation dimension caused by relying on single sensor data in existing poultry weight estimation technology.

[0004] To achieve the above purpose, the technical scheme adopted by the present application is as follows: In a first aspect, a poultry growth intelligent monitoring method based on multi-modal data fusion is provided, comprising: acquiring multi-modal data of poultry through multiple sensors; wherein the multi-modal data includes weight data, image data, activity data and environmental data; de-noising, de-duplicating and identifying the multi-modal data through a pre-processing algorithm to obtain standard data; establishing a multi-dimensional feature vector based on the standard data, and weighting and fusing the multi-dimensional feature vector through an attention mechanism to obtain fusion features; based on the fusion features, using a GMM-HMM model and a Transformer network to dynamically model and predict the growth trend of the poultry group, and calculating a group uniformity evaluation index to obtain an analysis result; generating a body weight estimate and a growth warning signal according to the analysis result.

[0005] Based on the above technical scheme, in the poultry growth intelligent monitoring method based on multi-modal data fusion provided in the application, by fusing weight, image, activity and environment multi-modal data and adopting attention mechanism for dynamic weighted fusion, the limitation of traditional methods relying on a single data source is effectively overcome, and the accuracy and robustness of poultry weight estimation are significantly improved; further based on the GMM-HMM model and the Transformer network, the group growth dynamics are modeled and predicted, realizing the leap from static weight estimation to dynamic trend analysis and early warning, so as to provide more comprehensive and timely growth state evaluation and risk warning support for farms, and improve the fine and intelligent level of breeding management.

[0006] In combination with the first aspect, in a possible implementation manner, the denoising, deduplication and identification of the multi-modal data through the preprocessing algorithm comprises: The weight data is denoised through an adaptive filtering algorithm to obtain denoised data; The multi-modal data is deduplicated based on RFID information and a timestamp; The image data is identified, segmented and feature-extracted through a target detection algorithm.

[0007] In combination with the first aspect, in a possible implementation manner, the identification, segmentation and feature extraction of the image data through the target detection algorithm comprises: A target detection model for poultry shape identification is constructed and trained; The image data is segmented and key point detected through the trained target detection model to obtain key point coordinates; The key point coordinates are converted into three-dimensional space coordinates based on camera calibration parameters to obtain three-dimensional coordinates; The shape parameters of the poultry are calculated according to the three-dimensional coordinates.

[0008] In combination with the first aspect, in a possible implementation manner, the target detection model further integrates a multi-scale key point attention mechanism; wherein the multi-scale key point attention mechanism comprises a scale hierarchical attention branch and a key point correlation attention module; The scale hierarchical attention branch comprises a plurality of scale attention sub-branches, which are used for reducing dimension of channels through standard convolution, extracting scale features through separable convolution, and generating scale attention weights through a sigmoid function; The key point correlation attention module is used for generating a core key point correlation matrix based on the anatomical structure of the poultry, calculating correlation attention weights through matrix multiplication, and performing weighted summation on the key point coordinates based on the correlation attention weights.

[0009] In conjunction with the first aspect above, in one possible implementation, the target detection model is optimized and trained using a composite loss function, which includes coordinate loss, structural constraint loss, and scale consistency constraint. The coordinate loss is the traditional mean square error loss, which is used to constrain the deviation between the key point coordinates and the labeled coordinates; The structural constraint loss is constructed based on anatomical structure rules and is used to constrain the relative distances between key points; the expression for the structural constraint loss is: In the formula, The first output of the object detection model i The key point and the first j Predicted distance between key points For the first i The key point and the first j The theoretical distance between these key points and the anatomical dimensions; The scale consistency constraint is used to constrain the consistency of the coordinates of the same keypoint predicted by feature layers at different scales; wherein, the expression of the scale consistency constraint is: In the formula, For the first l Key point prediction values ​​of layer features The mean of the predicted values ​​for all keypoints. L The number of attention sub-branches is the scale.

[0010] In conjunction with the first aspect above, in one possible implementation, the method for obtaining the fusion feature includes: The standard data is constructed into a graph structure; wherein the graph structure includes nodes and edges, the nodes represent weight features, morphological features, activity features and environmental features, and the edges represent potential associations between different feature data; The graph structure is input into the graph attention network, and cross-modal information propagation and fusion are performed by calculating the dynamic attention coefficients between nodes. The calculation of the dynamic attention coefficients introduces a conditional context vector composed of poultry growth stage identifiers. The conditional context vector is jointly encoded by the poultry's current growth stage and age information. Calculate the importance scores of the node features after updating through the graph attention network, and perform weighted pooling based on the importance scores to generate fused features.

[0011] In conjunction with the first aspect above, in one possible implementation, the use of the GMM-HMM model and Transformer network for poultry flock growth dynamic modeling and trend prediction includes: The hidden physiological state data of poultry is constructed by a GMM-HMM model, and the state transition probability matrix of the GMM-HMM model is modulated according to environmental data and group density; wherein the hidden physiological state data includes health sprint, stable maintenance, stress or sub-health and compensation growth; The encoded hidden physiological state data and the fusion features are taken as inputs by a Transformer network, and the growth curve and uniformity of the poultry group are predicted through a hierarchical attention mechanism; The meta-learning network is adopted to dynamically generate the fine-tuning amount of the parameters of the GMM-HMM model and the Transformer network according to the early data features of the new poultry in the shed, so as to realize the rapid self-adaptation of the model.

[0012] In combination with the first aspect, in a possible implementation manner, the expression of the group uniformity evaluation index is: ; Wherein, σ is the standard deviation of the body weight of the poultry, μ is the average body weight of the poultry; is the Shannon entropy of the state distribution of the poultry group, which is used to measure the chaos degree of the poultry group in the hidden physiological state distribution, n is any state in the hidden physiological state data, is the probability that the poultry is in state n occurs.

[0013] In combination with the first aspect, in a possible implementation manner, the generating of the body weight estimation and the growth warning signal according to the analysis result comprises: Based on the analysis result, the body weight-morphology joint estimation is performed through a regression algorithm to obtain the body weight estimation; When the growth curve deviates from a preset threshold or the uniformity evaluation index is abnormal, a growth warning signal is triggered.

[0014] Secondly, a poultry growth intelligent monitoring system based on multi-modal data fusion is provided, comprising: A multi-source data acquisition module comprising an intelligent weighing device, a 3D vision system, an RFID identification system and an environmental sensor array, for acquiring multi-modal data of poultry; A data preprocessing module for filtering, identifying and deduplicating the multi-modal data; A feature fusion module with a neural network based on an attention mechanism, for realizing weighted fusion of multi-modal data; A group modeling module with a GMM-HMM model and a Transformer network, for growth modeling and trend prediction; An intelligent decision module for generating weight estimation and early warning information.

[0015] In a third aspect, an electronic device is provided, comprising: a communication unit and a processing unit; the communication unit is configured to collect multi-modal data of poultry through a multi-sensor; the processing unit is configured to denoise, deduplicate and identify the multi-modal data through a preprocessing algorithm to obtain standard data; a multi-dimensional feature vector is established based on the standard data, the multi-dimensional feature vector is weighted and fused through an attention mechanism to obtain fused features; based on the fused features, a GMM-HMM model and a Transformer network are used for poultry group growth dynamic modeling and trend prediction, and a group uniformity evaluation index is calculated to obtain an analysis result; a weight estimation and a growth early warning signal are generated according to the analysis result.

[0016] In a fourth aspect, the present application provides an electronic device, comprising: a processor and a storage medium; the storage medium comprises instructions, and the processor is configured to execute the instructions to implement the method described in the first aspect and any possible implementation manner of the first aspect. The electronic device can be an electronic device, or a chip in the electronic device.

[0017] In a fifth aspect, the present application provides a poultry growth intelligent monitoring method system based on multi-modal data fusion, comprising: a multi-source data acquisition module, a data preprocessing module, a feature fusion module, a group modeling module and an intelligent decision module; wherein the multi-source data acquisition module is configured to collect multi-modal data of poultry; the data preprocessing module is configured to filter, identify and deduplicate the multi-modal data; the feature fusion module is configured to realize weighted fusion of the multi-modal data; the group modeling module is configured to perform growth modeling and trend prediction; and the intelligent decision module is configured to generate weight estimation and early warning information.

[0018] In a sixth aspect, the present application provides a computer readable storage medium, which stores instructions, and when the instructions are executed on an electronic device, the electronic device executes the method described in the first aspect and any possible implementation manner of the first aspect.

[0019] In a seventh aspect, the present application provides a computer program product comprising instructions, and when the computer program product is executed on an electronic device, the electronic device executes the method described in the first aspect and any possible implementation manner of the first aspect.

[0020] The application provides a poultry growth intelligent monitoring method and system based on multi-modal data fusion. By fusing multi-modal data and constructing an intelligent analysis model, high-precision, full-group continuous evaluation and abnormal early warning of the growth state of poultry groups are realized. The system can provide comprehensive evaluation covering multiple indicators such as body weight, uniformity, growth trend, etc., significantly improving the accuracy and comprehensiveness of monitoring; through dynamic modeling and prediction, early abnormality detection and risk warning are supported, providing real-time and objective data decision support for breeding process management, thereby effectively reducing the dependence on manpower and improving the intelligent level and economic benefits of poultry breeding management.

[0021] It should be understood that the description of technical features, technical solutions, beneficial effects or similar language in the present application does not imply that all features and advantages can be achieved in any single embodiment. On the contrary, it can be understood that the description of a feature or beneficial effect means that the specific technical feature, technical solution or beneficial effect is included in at least one embodiment. Therefore, the description of technical features, technical solutions or beneficial effects in the specification does not necessarily refer to the same embodiment. Further, the technical features, technical solutions and beneficial effects described in the embodiments can be combined in any appropriate manner. Those skilled in the art will understand that the embodiments can be implemented without one or more specific technical features, technical solutions or beneficial effects of the specific embodiments. In other embodiments, additional technical features and beneficial effects can be identified in specific embodiments that do not embody all embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 A system architecture diagram of a poultry growth intelligent monitoring system based on multi-modal data fusion provided by the embodiments of the present application; Figure 2 A flowchart of a poultry growth intelligent monitoring method based on multi-modal data fusion provided by the embodiments of the present application; Figure 3 A flowchart of another poultry growth intelligent monitoring method based on multi-modal data fusion provided by the embodiments of the present application; Figure 4 A structural schematic diagram of an electronic device provided by the embodiments of the present application; Figure 5 A hardware structural schematic diagram of an electronic device provided by the embodiments of the present application. DETAILED DESCRIPTION

[0023] In the description of the present application, unless otherwise specified, " / " means "or", for example, A / B can mean A or B. "And / or" herein is only a description of the relationship between the associated objects, which means that there can be three relationships, for example, A and / or B, which can mean: A alone, A and B exist at the same time, and B alone. In addition, "at least one" means one or more, and "multiple" means two or more. "First", "second", etc. do not limit the quantity and execution order, and "first", "second", etc. do not necessarily mean different.

[0024] It should be noted that in the present application, "exemplary" or "for example" is used to mean example, illustration or description. Any embodiment or design scheme described as "exemplary" or "for example" in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the use of "exemplary" or "for example" is intended to present the relevant concept in a specific manner.

[0025] The poultry growth intelligent monitoring method based on multi-modal data fusion provided by the embodiments of the present application can be applied to the poultry growth intelligent monitoring system 100 based on multi-modal data fusion as shown in Figure 1 As shown in Figure 1 The communication system includes a multi-source data acquisition module 10, a data preprocessing module 20, a feature fusion module 30, a group modeling module 40 and an intelligent decision module 50.

[0026] The multi-source data acquisition module 10 is configured to acquire multi-modal data of poultry. The data preprocessing module 20 is configured to filter, identify and remove duplicate processing of the multi-modal data. The feature fusion module 30 is configured to realize weighted fusion of the multi-modal data. The group modeling module 40 is configured to perform growth modeling and trend prediction. The intelligent decision module 50 is configured to generate weight estimation and early warning information.

[0027] To solve the technical problems of incomplete information dimension, large sample bias, poor dynamic adaptability and limited evaluation dimension caused by relying on single sensor data in the existing poultry weight estimation technology, the present application provides a poultry growth intelligent monitoring method based on multi-modal data fusion, which comprises the following steps: Acquiring multi-modal data of poultry through multiple sensors; wherein the multi-modal data includes weight data, image data, activity data and environmental data; Removing noise, removing duplicates and identifying the multi-modal data through a preprocessing algorithm to obtain standard data; A multi-dimensional feature vector is established based on standard data, and the multi-dimensional feature vector is weighted and fused through an attention mechanism to obtain fused features; Based on the fused features, a GMM-HMM model and a Transformer network are used to model and predict the growth dynamics of the poultry group, and a group uniformity evaluation index is calculated to obtain an analysis result. According to the analysis result, a weight estimation and a growth warning signal are generated.

[0028] Therefore, the technical problems of incomplete information dimension, large sample bias, poor dynamic adaptability, and limited evaluation dimension caused by the dependence of the existing poultry weight estimation technology on single sensor data are solved.

[0029] As shown in Figure 2 The poultry growth intelligent monitoring method based on multi-modal data fusion provided by the embodiments of the present application comprises: S201, multi-modal data of poultry is collected through multiple sensors.

[0030] The multi-modal data includes weight data, image data, activity data, and environmental data.

[0031] In some implementations, the collection method of the multi-modal data comprises: The weight data of the poultry is obtained through an intelligent weighing device; The image data of the poultry is collected through a 3D vision system; The activity frequency and duration of the poultry are recorded through an RFID identification system; The environmental data in the poultry house is monitored through an environmental sensor array; the environmental data includes temperature, humidity, and light intensity.

[0032] It should be noted that through the hardware cooperation and time synchronization mechanism, the continuous, correlated, and homologous collection of the weight, shape, behavior, and breeding environment of the poultry is realized for the first time, the problem of data fragmentation and space-time mismatch in traditional methods is solved, and a reliable data foundation is provided for subsequent high-precision fusion analysis, thereby ensuring the accuracy and reliability of the evaluation model from the source.

[0033] For example, the intelligent weighing device (precision ± 5 grams) deployed in the standardized stacked cage poultry house continuously obtains individual weight data when the chickens are feeding, and the 3D depth camera (resolution 1280x720) erected in the passageway synchronously collects the overhead and side view images of the chickens to extract body size parameters. The passive RFID tag worn on the leg of the chicken is scanned by the read-write device on both sides of the passageway in real time to record its entry and exit frequency and residence time. The environmental sensing nodes distributed at the four corners of the poultry house continuously monitor and report temperature, humidity and light intensity data. All sensor data are time-synchronized and packaged through a unified Internet of Things gateway to form a multi-modal data stream with a time stamp.

[0034] In S202, the multi-modal data is denoised, deduplicated and identified through a preprocessing algorithm to obtain standard data.

[0035] In some implementations, the denoising, deduplication and identification of the multi-modal data through the preprocessing algorithm comprises: The weight data is denoised through an adaptive filtering algorithm to obtain denoised data. The multi-modal data is deduplicated based on RFID information and a time stamp. The image data is identified, segmented and feature-extracted through a target detection algorithm.

[0036] It should be noted that through the preprocessing process for the characteristics of each modal data, the noise, redundancy and target mixing problems in the original data are systematically solved, the heterogeneous and original sensor output is converted into clean, regular and usable standard data for advanced analysis, laying a high-quality data foundation for subsequent feature fusion and modeling, and significantly improving the accuracy and reliability of the entire system analysis.

[0037] For example, a Kalman adaptive filter is used to real-time denoise the original weight data obtained by the intelligent weighing device, effectively filtering out signal jitter caused by poultry activity to obtain smooth weight time series data. Based on the RFID unique identifier of each poultry and the data collection time stamp, a spatio-temporal correlation rule is constructed to automatically identify and remove repeated records caused by a single poultry triggering the sensor multiple times in a short time, ensuring the uniqueness and representativeness of the data set. A YOLOv7 deep learning model trained on a large number of poultry images is used to perform real-time target detection and instance segmentation on the images collected by the 3D vision system, accurately separating the adhered or partially occluded poultry individuals, and extracting key morphological feature parameters such as body length and chest width. This technical solution.

[0038] In some implementations, the poultry identification, segmentation and feature extraction of the image data through the target detection algorithm comprises, as shown in Figure 3 ​S31. Construct a target detection model for poultry shape recognition and train it; S32. Perform image segmentation and key point detection on the image data through the trained target detection model to obtain key point coordinates; S33. Perform three-dimensional space coordinate conversion on the key point coordinates based on the camera calibration parameters to obtain three-dimensional coordinates; S34. Calculate the shape parameters of the poultry according to the three-dimensional coordinates.

[0039] It should be noted that the shape parameters include body length, chest width, and tibia length, the calculation method of the body length is to take the vector from the head key point to the torso center key point, and the modulus of the sum of the vector from the torso center key point to the tail key point as the body length of the poultry; the calculation method of the chest width is to take the torso center key point as the reference in the point cloud data of the poultry torso region, fit an ellipse on the plane perpendicular to the body length direction, and take the short axis length of the ellipse as the chest width estimate; and the calculation method of the tibia length is to estimate the tibia length through a regression model according to the spatial distance relationship between the foot key point and the torso key point, combined with the prior knowledge of poultry breeds.

[0040] It should be noted that this technical solution realizes automatic, high-precision, and non-contact measurement from two-dimensional images to three-dimensional shape parameters, not only overcoming the shortcomings of traditional manual measurement, such as low efficiency and easy to cause stress to poultry, but also providing reliable and quantitative data input for subsequent integration of morphological indicators into growth evaluation models.

[0041] For example, the target detection algorithm is based on the HRNet deep learning framework and is trained on a labeled image dataset containing multiple breeds and multiple poses of poultry, forming a special model that can output instance segmentation masks and 24 anatomical key points (such as beak tip, head top, wing root, knee joint, etc.) at the same time; during processing, the model inferences each input image and outputs the contour mask of each poultry and the two-dimensional pixel coordinates of each key point; then, the system uses the internal and external parameters of the binocular camera obtained in advance through the checkerboard calibration method, combined with the triangulation principle and the RANSAC optimization algorithm, to convert the corresponding two-dimensional key point pairs into accurate three-dimensional space coordinates; finally, based on this three-dimensional point cloud data, the system automatically calculates a series of shape parameters of the poultry, such as body length (occipital bone to tail vertebra distance), chest width (left and right shoulder joint distance), tibia length (tarsal joint to toe distance), etc.

[0042] In some implementations, the target detection model also integrates a multi-scale key point attention mechanism; wherein the multi-scale key point attention mechanism includes a scale hierarchical attention branch and a key point correlation attention module. The scale hierarchical attention branch includes several scale attention sub-branches, which are used for dimension reduction processing of channels by standard convolution, extraction of scale features by separable convolution, and generation of scale attention weights by a sigmoid function. The key point correlation attention module is used for generating a core key point correlation matrix based on the anatomical structure of poultry, calculating correlation attention weights by matrix multiplication, and performing weighted summation on key point coordinates based on the correlation attention weights.

[0043] It should be noted that by combining anatomical prior knowledge with adaptive multi-scale learning, the robustness and positioning accuracy of the model for poultry key point detection in complex farming scenarios (such as occlusion, crowding, and variable posture) are effectively improved, thereby providing a more stable and reliable data foundation for subsequent three-dimensional reconstruction and morphological parameter calculation.

[0044] For example, the target detection model uses HRNet as the backbone network and integrates a multi-scale key point attention mechanism. The mechanism includes a scale hierarchical attention branch and a key point correlation attention module: the former extracts scale features under different receptive fields from the feature map through three parallel depth separable convolution layers with expansion rates of 1, 3, and 6, respectively, generates an attention map through a Sigmoid function, and adaptively enhances the feature response to the overall outline and local details (such as the head and legs) of the poultry; the latter constructs a learnable sparse correlation matrix based on the anatomical prior of the spine-torso-limb of poultry, predefines strong correlation groups such as "beak-crown-neck vertebra" and "wing root-thoracic vertebra-knee joint" in the matrix, calculates the correlation weights between the features of each key point through matrix multiplication, and performs weighted fusion on the features, so that the model can significantly refer to the feature information of its anatomically related points (such as the wing root and tarsal joint) when predicting a key point (such as the knee joint).

[0045] In some implementations, the target detection model is optimized and trained using a composite loss function, which includes a coordinate loss, a structure constraint loss, and a scale consistency constraint; The coordinate loss is a traditional mean square error loss, which is used to constrain the deviation of the key point coordinates from the labeled coordinates; The structure constraint loss is constructed based on anatomical structure rules and is used to constrain the relative distances between key points; wherein the expression of the structure constraint loss is: ; in the formula, is the predicted distance between the i th key point and the j th key point output by the target detection model, is the theoretical distance between the i th key point and the j th key point in anatomy; The scale consistency constraint is used to constrain the consistency of the same key point coordinates predicted by different scale feature layers. ; in the formula, is the key point prediction value of the first l layer feature, is the mean value of all key point prediction values, L is the number of scale attention sub-branches.

[0046] It should be noted that by embedding anatomical structure knowledge as a soft constraint into the loss function and forcing the consistency of multi-scale prediction, the model is effectively guided to learn a key point representation that conforms to biological morphology and is scale-robust, significantly improving the prediction accuracy and structural rationality of the model in difficult scenarios such as partial occlusion, severe congestion, or non-standard poses, providing a higher fidelity two-dimensional prediction basis for subsequent three-dimensional reconstruction.

[0047] For example, the mean square error loss is used as the basis for the coordinate loss term; further introduce the structure constraint loss term, which according to the standard poultry anatomical atlas, predefines the theoretical length range of key bone segments such as "neck length" and "trunk length", and constrains the length of the corresponding line segment composed of the predicted key points to conform to the theoretical range in the loss calculation; At the same time, the scale consistency constraint term is added, which calculates the variance between the same key point coordinates predicted by different scale attention branches (such as three branches corresponding to original Figure 1 / 4, 1 / 8, 1 / 16 down-sampling rate) in the model, forcing the features at different receptive fields to reach a consensus on the positioning of the key points.

[0048] S203, establish a multi-dimensional feature vector based on standard data, and obtain a fused feature by weighting and fusing the multi-dimensional feature vector through an attention mechanism.

[0049] In some implementations, the fused feature is obtained in the following manner: The standard data is constructed as a graph structure; wherein the graph structure includes nodes and edges, the nodes represent weight features, morphology features, activity features, and environment features, and the edges represent potential associations between different feature data; The graph structure is input into a graph attention network to perform cross-modal information propagation and fusion by calculating dynamic attention coefficients between nodes; wherein the calculation of the dynamic attention coefficients introduces a conditional context vector composed of a poultry growth stage identifier, and the conditional context vector is encoded by the current growth stage and age information of the poultry; The importance score of the node feature updated by the graph attention network is calculated, and the fused feature is generated by weighted pooling according to the importance score.

[0050] It should be noted that, unlike traditional GAT, by introducing a conditional context vector, which is encoded by the current growth stage identifier of the poultry (classified by a small network from morphological parameters) and time information (such as age), the attention mechanism can dynamically adjust the importance between modalities according to the growth stage of the chicken. For example, in the brooding period, the influence weight of the environmental temperature may be greater; in the fattening period, the weight and the association weight of the morphology may be higher.

[0051] It should be noted that, by introducing a graph structure to represent multi-modal data and using a conditional graph attention mechanism for fusion, the scheme is no longer a simple feature stacking or linear weighting, but realizes the reasoning and learning of complex nonlinear relationships between modalities. This method can more intelligently capture deep relationships such as "how does the decrease in poultry activity under high temperature environment affect its weight gain efficiency", thereby generating more information-rich and more effective fusion features for downstream tasks, significantly improving the performance upper limit of the entire system.

[0052] For example, first, the weight time series features (such as daily weight gain), multi-dimensional morphological features (such as body length, chest width), activity features (such as daily activity frequency), and environmental features (such as temperature and humidity index) obtained after preprocessing are constructed into a heterogeneous graph, where each feature is a node, and the edges between nodes are initialized as a preset weight based on the correlation between feature types (such as morphology and weight, activity and environment). Subsequently, the graph is input into a two-layer graph attention network, which introduces a conditional context vector generated by embedding the current growth stage of the poultry (such as brooding period, growth period) and the specific age, which dynamically modulates the attention calculation through a learnable feedforward network, so that the environmental features are given higher attention in the brooding period, while the association between morphological and weight features is strengthened in the growth period. Finally, the network outputs a fixed-dimensional, deeply fused feature vector through an adaptive weighted pooling layer that integrates the updated features of each node and their importance scores.

[0053] S204, based on the fusion features, using GMM-HMM model and Transformer network for poultry population growth dynamic modeling and trend prediction, and calculating population uniformity evaluation index to obtain analysis results.

[0054] In some implementations, the GMM-HMM model and the Transformer network are used to model the growth dynamics of the poultry population and predict trends, including: The GMM-HMM model is used to construct hidden physiological state data for the poultry, and the state transition probability matrix of the GMM-HMM model is modulated according to the environmental data and the population density; wherein the hidden physiological state data includes health sprint, stable maintenance, stress or sub-health, and compensation growth; The encoded hidden physiological state data and the fusion features are taken as inputs by the Transformer network, and the growth curve and uniformity of the poultry population are predicted through a hierarchical attention mechanism. The meta-learning network is used to dynamically generate the fine-tuning amount of the GMM-HMM model and the Transformer network parameters according to the early data features of the new poultry, so as to realize the rapid adaptation of the model.

[0055] It should be noted that by combining the environment modulated HMM, the time series Transformer and the meta-learning adaptive three, an intelligent growth modeling system is constructed, which can dynamically reflect the influence of breeding conditions, accurately predict long-term trends, and quickly migrate to new production batches, thereby fundamentally solving the problems of large prediction deviation and poor generalization ability of traditional static models, and providing core algorithm support for realizing truly predictive breeding management.

[0056] For example, first, a four-state (healthy sprint, stable maintenance, stress / sub-health, compensation growth) GMM-HMM model is used, and the state transition probability matrix is not fixed, but is dynamically adjusted by a lightweight modulation network according to real-time environmental temperature, humidity and population density calculated based on RFID data, thereby generating a hidden physiological state sequence matched with the current breeding conditions; then, the state sequence and the aforementioned fusion features are encoded and input into a Transformer network with an encoder-decoder structure, the decoder of which uses a hierarchical attention mechanism to focus on short-term growth fluctuations at the bottom layer and capture long-term trends at the high layer, and finally outputs the prediction curve of the population average weight and the uniformity (CU value) change trend in the future M days; in order to adapt to different varieties or batches of poultry, a meta-learning network based on the model-agnostic meta-learning (MAML) framework is also introduced, which only needs to analyze the early multi-modal data features of the new poultry in the first week, and can generate a small-scale and targeted gradient update amount for the key parameters of the above GMM-HMM modulation network and Transformer network, so that the basic model can be quickly fine-tuned on a small amount of new data to adapt to the new population.

[0057] In some implementations, the expression of the population uniformity evaluation index is: ; wherein, σ is the standard deviation of the weight of the poultry, μ is the average weight of the poultry; is the Shannon entropy of the state distribution of the poultry population, which is used to measure the degree of chaos of the poultry population in the hidden physiological state distribution, n is any state in the hidden physiological state data, To be in condition for poultry n Probability of occurrence.

[0058] It should be noted that by combining the statistical weight dispersion with the information entropy measure of physiological state distribution disorder, the comprehensive quantitative evaluation of the "apparent weight uniformity" and "intrinsic physiological state uniformity" of the group is creatively realized, so that the index can not only reflect the weight difference, but also sensitively capture the group state differentiation caused by disease latency, environmental stress and other factors, which has not yet been reflected in the weight, providing the grower with an earlier and more in-depth group health and growth consistency warning signal than traditional methods.

[0059] For example, first, based on the hidden physiological state sequence inferred by the GMM-HMM model, the proportion of individuals in each state (health sprint, stable maintenance, stress / sub-health, compensation growth) of the entire poultry group is calculated as a probability distribution to calculate the Shannon entropy H(N), which quantifies the uniformity of the group in physiological state distribution; At the same time, the system calculates the standard deviation σ and the average value μ of the current batch of poultry weight, and obtains the traditional weight uniformity component (1-σ / μ); Finally, a comprehensive group uniformity evaluation index is generated by fusing the Shannon entropy formula.

[0060] S205, generating weight estimation and growth warning signal according to analysis result.

[0061] In some implementations, the generating weight estimation and growth warning signal according to analysis result comprises: Based on the analysis result, the weight-morphology joint estimation is carried out by regression algorithm to obtain the weight estimation; When the growth curve deviates from the preset threshold or the uniformity evaluation index is abnormal, the growth warning signal is triggered.

[0062] It should be noted that by combining the data-driven regression model with the intelligent rule judgment based on dynamic threshold, not only a more stable and anti-interference weight estimation than single weighing is provided, but more importantly, an objective, timely and hierarchical response automatic warning mechanism is established, so that the manager can take intervention measures when the growth curve deviates significantly or the group uniformity deteriorates at the early stage, thereby transforming the traditional lagging management relying on experience observation into precise and preventive management driven by data.

[0063] Exemplarily, first, morphological features (such as body length and chest width) contained in the analysis result and historical weight data are taken as inputs, a gradient boosting regression tree algorithm combined with feature cross terms is used for training and inference, morphological-based weight accurate estimation is realized, and final individual and group weight estimates are output; at the same time, the system compares the predicted growth curve in the analysis result with the preset dynamic threshold band (the threshold band is generated based on the historical optimal growth trajectory quantile under the same breed and breeding mode), monitors whether the comprehensive uniformity evaluation index is continuously lower than the preset warning value for multiple periods, and when any condition is met, the system immediately generates a growth warning signal, which is divided into three levels of "attention" (yellow), "warning" (orange) and "alarm" (red) according to the deviation degree, and is synchronously prompted to the breeding manager through a visual interface and message pushing.

[0064] Based on the above technical solution, the poultry growth intelligent monitoring method based on multi-modal data fusion provided in the application realizes high-precision, full-group continuous evaluation and abnormal early warning of the growth state of the poultry group by fusing multi-modal data and constructing an intelligent analysis model. The system can provide comprehensive evaluation of multi-dimensional indexes such as weight, uniformity, growth trend, etc., significantly improving the accuracy and comprehensiveness of monitoring; through dynamic modeling and prediction, early abnormality detection and risk warning are supported, providing real-time and objective data decision support for breeding process management, thereby effectively reducing the dependence on manpower and improving the intelligent level and economic benefits of poultry breeding management.

[0065] In a possible implementation manner, the embodiment of the application further provides a poultry growth intelligent monitoring system based on multi-modal data fusion, comprising: A multi-source data acquisition module comprising an intelligent weighing device, a 3D vision system, an RFID identification system and an environmental sensor array, for acquiring multi-modal data of poultry; A data preprocessing module for filtering, identifying and deduplicating the multi-modal data; A feature fusion module with a neural network based on an attention mechanism for realizing weighted fusion of multi-modal data; A group modeling module with a GMM-HMM model and a Transformer network for growth modeling and trend prediction; An intelligent decision-making module for generating weight estimates and warning information.

[0066] Illustratively, the poultry growth intelligent monitoring system based on multi-modal data fusion, the multi-source data acquisition module is jointly constituted by the intelligent electronic scale deployed at the outlet of the breeding cage, the binocular 3D depth camera installed above the channel, the RFID reader and writer distributed near the feeding trough and water line, and the Internet of Things environment sensing node deployed in the poultry house, realizing synchronous acquisition and uploading of multi-source data; the data preprocessing module runs in the edge computing gateway, adopts Kalman filtering, instance segmentation based on YOLOv7, and RFID deduplication algorithm based on space-time rules, and performs regularization on the original data; the feature fusion module is deployed in the cloud server, adopts a fusion model constructed with a graph attention network as the core, and deeply fuses the weight, shape, activity and environment features after preprocessing; the group modeling module also runs in the cloud, realizes growth dynamic modeling and trend prediction through the cascade of the GMM-HMM model modulated by the environmental conditions and the hierarchical attention Transformer network; the intelligent decision module comprehensively considers the above results, adopts gradient boosting regression tree for weight evaluation, and generates graded early warning according to the comparison result with the dynamic threshold.

[0067] Based on the above technical solutions, the full-link automation from data perception, cleaning, fusion, modeling to decision output is realized, discrete multi-sensor information is converted into continuous and interpretable growth state evaluation and risk warning, an "end-edge-cloud" collaborative integrated intelligent solution is provided for the farm, and the real-time of monitoring, the accuracy of evaluation and the predictability of management are significantly improved.

[0068] The above mainly introduces the scheme of the embodiments of the present application from the perspective of device implementation. It can be understood that each device, for example, an electronic device, includes at least one of a corresponding hardware structure and a software module for executing each function. Those skilled in the art should easily realize that, in combination with the units and algorithm steps of the examples described in the embodiments disclosed herein, the present application can be realized in the form of hardware or a combination of hardware and computer software. Whether a certain function is driven by hardware or computer software, it depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0069] The embodiments of the present application can divide the functional units of the electronic device according to the above method examples, for example, each functional unit can be divided according to each function, or two or more functions can be integrated in one processing unit. The integrated unit can be realized in the form of hardware or software functional unit. It should be noted that the division of units in the embodiments of the present application is illustrative, and is only a logical function division. When actually implemented, there can be another division method.

[0070] In the case of employing an integrated unit, Figure 4 A possible structural diagram of an electronic device (denoted as electronic device 50) involved in the above embodiments is shown, which includes a processing unit 501 and a communication unit 502, and can further include a storage unit 503. Figure 4 The shown structural diagram can be used to illustrate the structure of the electronic device involved in the above embodiments.

[0071] When Figure 4 When the shown structural diagram is used to illustrate the structure of the electronic device involved in the above embodiments, the processing unit 501 is used to control and manage the actions of the electronic device, the communication unit 502 is used for the electronic device to communicate with other devices, and the storage unit 503 is used to store the program code and data of the electronic device.

[0072] For example, the communication unit 502 is used to collect multi-modal data of poultry through multiple sensors; The processing unit 501 is used to denoise, deduplicate and identify the multi-modal data through a preprocessing algorithm to obtain standard data; a multi-dimensional feature vector is established based on the standard data, and the multi-dimensional feature vector is weighted and fused through an attention mechanism to obtain fusion features; based on the fusion features, GMM-HMM model and Transformer network are used for poultry group growth dynamic modeling and trend prediction, and a group uniformity evaluation index is calculated to obtain analysis results; and a weight estimate and a growth warning signal are generated according to the analysis results.

[0073] Among them, the processing unit 501 can be a processor or a controller, and the communication unit 502 can be a communication interface, a transceiver, a transceiver, a transceiver circuit, a transceiver device, etc. Among them, the communication interface is a general term and can include one or more interfaces. The storage unit 503 can be a memory. When the electronic device 50 is a chip, the processing unit 501 can be a processor or a controller, and the communication unit 502 can be an input interface and / or an output interface, a pin or a circuit, etc. The storage unit 503 can be a storage unit (e.g., register, cache, etc.) within the chip, or can be a storage unit (e.g., read-only memory (ROM), random access memory (RAM), etc.) located outside the chip.

[0074] The communication unit can also be referred to as a transceiver unit. The antenna and the control circuit with transceiving functions in the electronic device 50 can be regarded as a communication unit 502 of the electronic device 50, and the processor with processing functions can be regarded as a processing unit 501 of the electronic device 50. Optionally, the device for realizing the receiving function in the communication unit 502 can be regarded as a communication unit, which is used to perform the receiving steps in the embodiments of the present application, and the communication unit can be a receiver, a receiver, a receiving circuit, etc. The device for realizing the sending function in the communication unit 502 can be regarded as a sending unit, which is used to perform the sending steps in the embodiments of the present application, and the sending unit can be a transmitter, a sender, a sending circuit, etc.

[0075] Figure 4 The integrated units in the above embodiments can be stored in a computer readable storage medium if they are realized in the form of software function modules and sold or used as independent products. Based on such understanding, the technical solutions of the embodiments of the present application essentially or the parts that make contributions to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods described in the embodiments of the present application. The storage medium storing the computer software product includes a U disk, a mobile hard disk, a read-only memory, a random access memory, a magnetic disk or an optical disk, and various media that can store program codes.

[0076] Figure 4 The units in the above embodiments can also be referred to as modules, for example, the processing unit can be referred to as a processing module.

[0077] The embodiments of the present application also provide a hardware structure diagram of an electronic device (denoted as an electronic device 60), which is shown in Figure 5 The electronic device 60 includes a processor 601, and optionally further includes a memory 602 connected with the processor 601.

[0078] In a first possible implementation, referring to Figure 5 The electronic device 60 further includes a transceiver 603. The processor 601, the memory 602 and the transceiver 603 are connected through a bus. The transceiver 603 is used for communicating with other devices or communication networks. Optionally, the transceiver 603 can include a transmitter and a receiver. The device for realizing the receiving function in the transceiver 603 can be regarded as a receiver, which is used to perform the receiving steps in the embodiments of the present application. The device for realizing the sending function in the transceiver 603 can be regarded as a transmitter, which is used to perform the sending steps in the embodiments of the present application.

[0079] Based on the first possible implementation, Figure 5 The structural schematic diagram shown can be used to show the structure of the electronic device involved in the above embodiments.

[0080] Among them, Figure 5 The system chip in the electronic device can also be shown. In this case, the actions performed by the electronic device described above can be implemented by the system chip, and the specific actions performed can be referred to above and will not be described here.

[0081] In the implementation process, each step in the method provided by the embodiment can be completed by the integrated logic circuit of hardware in the processor or the instruction in the form of software. The steps of the method disclosed by the embodiment of the present application can be directly embodied as hardware processor execution completion, or executed by the combination of hardware and software modules in the processor.

[0082] The processor in the present application can include but is not limited to at least one of the following: central processing unit (CPU), microprocessor, digital signal processor (DSP), microcontroller unit (MCU), or artificial intelligence processor and various types of computing devices running software, each of which can include one or more cores for executing software instructions to perform operations or processing. The processor can be a separate semiconductor chip, or can be integrated with other circuits as a semiconductor chip, for example, it can constitute a SoC (system on chip) with other circuits (such as coding and decoding circuits, hardware acceleration circuits or various buses and interface circuits), or it can be integrated as an internal processor in the ASIC, and the ASIC integrated with the processor can be packaged separately or packaged together with other circuits. In addition to including cores for executing software instructions to perform operations or processing, the processor can further include necessary hardware accelerators, such as field programmable gate array (FPGA), PLD (programmable logic device), or logic circuits implementing special logic operations.

[0083] The memory in the embodiments of the present application can include at least one of the following types: read-only memory (ROM) or other types of static storage devices that can store static information and instructions, random access memory (RAM) or other types of dynamic storage devices that can store information and instructions, and can also be electrically erasable programmable read-only memory (EEPROM). In some scenarios, the memory can also be a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), magnetic disk storage medium or other magnetic storage device, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and capable of being accessed by a computer, but not limited to this.

[0084] The embodiments of the present application also provide a computer readable storage medium including instructions, which, when executed on a computer, cause the computer to perform any of the above methods.

[0085] The embodiments of the present application also provide a computer program product including instructions, which, when executed on a computer, cause the computer to perform any of the above methods.

[0086] The embodiments of the present application also provide a chip, which includes a processor and an interface circuit, the interface circuit is coupled with the processor, the processor is used to run a computer program or instructions to implement the above method, and the interface circuit is used to communicate with other modules outside the chip.

[0087] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present application are generated. The computer can be a general purpose computer, a special purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another computer readable storage medium, for example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (digital subscriber line, DSL)) or wireless (such as infrared, wireless, microwave, etc.) mode. The computer readable storage medium can be any available medium that can be accessed by a computer or data storage device including one or more servers, data centers, etc. integrated with the medium. The available medium can be magnetic medium (such as floppy disk, hard disk, magnetic tape), optical medium (such as DVD), or semiconductor medium (such as solid state disk (solid state disk, SSD)) and the like.

[0088] Although the present application is described herein in conjunction with various embodiments, other variations and modifications of the disclosed embodiments can be understood and implemented by those skilled in the art upon review of the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and the indefinite article "a" or "an" does not exclude a plurality. A single processor or other unit can implement several functions listed in the claims. Measures described in mutually different dependent claims can be combined and implemented.

[0089] Although the present application is described herein in conjunction with various embodiments, other variations and modifications of the disclosed embodiments can be understood and implemented by those skilled in the art upon review of the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and the indefinite article "a" or "an" does not exclude a plurality. A single processor or other unit can implement several functions listed in the claims. Measures described in mutually different dependent claims can be combined and implemented.

Claims

1. A method for intelligent monitoring of poultry growth based on multimodal data fusion, characterized in that, include: Multimodal data of poultry is collected using multiple sensors; wherein the multimodal data includes weight data, image data, activity data, and environmental data. Preprocessing algorithms are used to denoise, deduplicatize, and identify multimodal data to obtain standard data; A multidimensional feature vector is established based on standard data, and the multidimensional feature vector is weighted and fused through an attention mechanism to obtain the fused feature. Based on the fusion features, the GMM-HMM model and Transformer network were used to model the growth dynamics and predict the trend of poultry flocks, and the flock evenness evaluation index was calculated to obtain the analysis results. Based on the analysis results, weight estimates and growth warning signals are generated.

2. The intelligent monitoring method for poultry growth based on multimodal data fusion according to claim 1, characterized in that, The preprocessing algorithm for denoising, deduplication, and identification of multimodal data includes: The weight data is denoised using an adaptive filtering algorithm to obtain denoised data. Deduplication of multimodal data is performed based on RFID information and timestamps; Poultry identification, segmentation, and feature extraction are performed on image data using object detection algorithms.

3. The intelligent monitoring method for poultry growth based on multimodal data fusion according to claim 2, characterized in that, The process of identifying, segmenting, and extracting features from image data using a target detection algorithm includes: Construct and train a target detection model for poultry morphology recognition; The trained object detection model is used to perform image segmentation and key point detection on the image data to obtain the key point coordinates. Based on the camera calibration parameters, the coordinates of the key points are transformed into three-dimensional spatial coordinates to obtain three-dimensional coordinates; The morphological parameters of poultry are calculated based on three-dimensional coordinates.

4. The intelligent monitoring method for poultry growth based on multimodal data fusion according to claim 3, characterized in that, The target detection model also integrates a multi-scale keypoint attention mechanism; wherein, the multi-scale keypoint attention mechanism includes a scale-layered attention branch and a keypoint association attention module; The scale-level attention branch includes several scale attention sub-branches, which are used to reduce the dimensionality of the channels through standard convolution, extract scale features through separable convolution, and generate scale attention weights through the sigmoid function. The key point association attention module is used to generate a core key point association matrix based on the anatomical structure of poultry, calculate the association attention weights through matrix multiplication, and perform a weighted summation of the key point coordinates based on the association attention weights.

5. The intelligent monitoring method for poultry growth based on multimodal data fusion according to claim 4, characterized in that, The target detection model is optimized and trained using a composite loss function, which includes coordinate loss, structural constraint loss, and scale consistency constraint. The coordinate loss is the traditional mean square error loss, which is used to constrain the deviation between the key point coordinates and the labeled coordinates; The structural constraint loss is constructed based on anatomical structure rules and is used to constrain the relative distances between key points; the expression for the structural constraint loss is: In the formula, The first output of the object detection model i The key point and the first j Predicted distance between key points For the first i The key point and the first j The theoretical distance between these key points and the anatomical structure; The scale consistency constraint is used to constrain the consistency of the coordinates of the same keypoint predicted by feature layers at different scales; wherein, the expression of the scale consistency constraint is: In the formula, For the first l Key point prediction values ​​of layer features The mean of the predicted values ​​for all key points. L The number of attention sub-branches is the scale.

6. The intelligent monitoring method for poultry growth based on multimodal data fusion according to claim 1, characterized in that, The methods for obtaining the fusion features include: The standard data is constructed into a graph structure; wherein the graph structure includes nodes and edges, the nodes represent weight features, morphological features, activity features and environmental features, and the edges represent potential associations between different feature data; The graph structure is input into the graph attention network, and cross-modal information propagation and fusion are performed by calculating the dynamic attention coefficients between nodes. The calculation of the dynamic attention coefficients introduces a conditional context vector composed of poultry growth stage identifiers. The conditional context vector is jointly encoded by the poultry's current growth stage and age information. Calculate the importance scores of the node features after updating through the graph attention network, and perform weighted pooling based on the importance scores to generate fused features.

7. The intelligent monitoring method for poultry growth based on multimodal data fusion according to claim 1, characterized in that, The method of using GMM-HMM model and Transformer network for poultry flock growth dynamic modeling and trend prediction includes: Hidden physiological state data for poultry were constructed using the GMM-HMM model, and the state transition probability matrix of the GMM-HMM model was modulated based on environmental data and population density. The hidden physiological state data included healthy sprint, stable maintenance, stress or sub-health, and compensatory growth. Using a Transformer network, encoded hidden physiological state data and fused features are used as inputs to predict the growth curve and evenness of poultry flocks through a hierarchical attention mechanism. A meta-learning network is used to dynamically generate fine-tuning parameters for the GMM-HMM model and Transformer network based on early data characteristics of newly admitted poultry, enabling the model to adapt quickly.

8. The intelligent monitoring method for poultry growth based on multimodal data fusion according to claim 7, characterized in that, The expression for the population evenness evaluation index is: ； in, σ The standard deviation of poultry body weight. μ The average weight of poultry; Shannon entropy, representing the distribution of states in a poultry population, is used to measure the degree of disorder in the distribution of hidden physiological states within a poultry population. n To hide any state in the physiological state data, For poultry to be in condition n The probability of occurrence.

9. The intelligent monitoring method for poultry growth based on multimodal data fusion according to claim 7, characterized in that, The generation of weight estimation and growth early warning signals based on the analysis results includes: Based on the analysis results, a weight-morphology joint estimation was performed using a regression algorithm to obtain the weight estimate; When the growth curve deviates from the preset threshold or the uniformity evaluation index is abnormal, a growth warning signal is triggered.

10. A poultry growth intelligent monitoring system based on multimodal data fusion according to claim 1, characterized in that, include: The multi-source data acquisition module includes an intelligent weighing device, a 3D vision system, an RFID identification system, and an environmental sensor array, used to collect multimodal data of poultry; The data preprocessing module is used to filter, identify, and deduplicatize multimodal data; The feature fusion module has a built-in attention-based neural network for weighted fusion of multimodal data. The population modeling module, with its built-in GMM-HMM model and Transformer network, is used for growth modeling and trend prediction. The intelligent decision-making module is used to generate weight estimates and early warning information.

Images