Poultry weight cleaning and uniformity evaluation method and system for farming decision-making

Abstract

The present application relates to the poultry breeding weighing technology field, especially is the poultry weight cleaning and uniformity evaluation method and system for breeding decision. The poultry weight data is collected in real time through the intelligent weighing sensor array, and the data is screened based on the fluctuation threshold weight strategy and the reference weight proportion range, the day age linear growth model, the growth rule constraint, the preliminary effective sample set is formed; the density peak value area of the weight data is identified by the mean shift clustering algorithm combined with the adaptive kernel bandwidth, and the outliers are automatically removed; the high order moment characteristics of the weight distribution are statistically distinguished by the skewness-kurtosis low-dimensional embedding domain, and the distribution result is output. When the detection data form is positively skewed, the square root transformation and mean correction are performed on the weight data, the high-precision poultry weight statistical index and uniformity evaluation result are generated, and the growth trend curve is drawn in real time, which can provide reliable data support for feeding strategy adjustment, health monitoring and culling decision.

Description

Technical Field

[0001] This invention relates to the field of poultry farming weighing technology, and in particular to a method and system for assessing poultry weight cleaning and uniformity for farming decision-making. Background Technology

[0002] In modern large-scale poultry farms, poultry weight is a core indicator for measuring growth status, evaluating feeding effectiveness, and formulating slaughter plans. However, the complex environment of farms, frequent poultry activity, and stress behaviors make weighing equipment susceptible to factors such as vibration, electromagnetic interference, or platform offset. This results in real-time weight data often containing significant noise, outliers, and distorted distributions. Traditional poultry weight collection and analysis methods typically rely on single-point weighing devices and use fixed thresholds, simple averages, or empirical rules as data processing tools. This approach is insufficient to meet the practical needs of today's large-scale, real-time, and automated poultry farming.

[0003] Existing methods for processing poultry weight data often employ static threshold rules for anomaly detection, such as setting fixed minimum and maximum weight ranges to remove obviously invalid measurements. However, due to significant differences in growth rates among poultry at different ages, fixed thresholds cannot adapt to or satisfy the constantly changing weight distribution, easily leading to the accidental deletion of a large amount of normal data or the failure to remove abnormal data in a timely manner. Furthermore, traditional methods generally lack the ability to deeply process fluctuations in continuously collected data. For example, continuous micro-shaking of 1-5g caused by poultry's instability will generate a large amount of duplicate information in the original weight data, and simple deduplication rules cannot accurately identify the repetitive dynamic features. In addition, traditional data processing workflows are mostly linear, single-stage processing modes, lacking hierarchical cleaning strategies and edge computing mechanisms, making it impossible to achieve real-time and efficient processing in high-frequency acquisition scenarios. Their large processing latency and limited throughput make it difficult to guarantee the real-time, reliability, and robustness of weight data in large-scale farming decision-making scenarios, easily interfering with poultry farming decisions and leading to a significant decline in overall poultry growth indicators and farming quality. Summary of the Invention

[0004] This invention overcomes the shortcomings of the prior art and provides a method and system for assessing poultry weight cleaning and uniformity for breeding decisions.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] The first aspect of this invention provides a method for assessing poultry weight washing and uniformity for breeding decision-making, comprising the following steps:

[0007] S101: Collect real-time weight data of poultry individuals in the target breeding area through an intelligent weighing sensor array, detect duplicate data of real-time weight data by setting a fluctuation threshold and deduplicating the data, and after deduplication, filter invalid data of weight data based on the reference weight and corresponding proportion range of poultry individuals and the linear growth model of weight by age to obtain a preliminary set of weight data samples.

[0008] S102: If the number of samples in the initially screened weight data sample set is greater than the preset threshold, then proceed to the mean drift clustering calculation program.

[0009] S103: Upload the initially screened weight data sample set to the cloud platform. Based on the local density distribution and density clustering characteristics of the weight data sample set, preset the optimal kernel bandwidth and radial symmetric kernel function. Calculate the mean drift clustering of the weight dataset using the optimal kernel bandwidth and radial symmetric kernel function. Finally, remove outlier data points with weighing errors based on the mean drift clustering results to obtain the effective weight dataset of poultry cluster weighing.

[0010] S104: Establish a low-dimensional embedding feature domain for the skewness-kurtosis plane, calculate the distribution shape of the effective weight dataset, obtain the distribution skewness and distribution kurtosis, embed the higher-order moment feature vector of the effective weight data into the low-dimensional embedding feature domain based on the distribution skewness and distribution kurtosis, perform threshold judgment analysis of statistical distribution features, and output the analysis results of one type of statistical distribution features and the analysis results of two types of statistical distribution features.

[0011] S105: If the results are displayed as a type II statistical distribution characteristic analysis, the effective weight data is transformed by square root, and a robust mean correction calculation is performed on the transformed effective weight data. The output is the weight statistical index and uniformity assessment value of the poultry for accurate weighing, and a trend line of poultry growth dynamics is plotted to make adjustments to breeding decisions.

[0012] Preferably, step S101 specifically includes the following steps:

[0013] The target breeding area is identified, and the weight of individual poultry within the target breeding area is collected in real time through an intelligent weighing sensor array to obtain real-time weight data of different poultry individuals.

[0014] A pre-set fluctuation threshold is used to calculate the fluctuation index weighted moving sequence of real-time weight data in the processing time sequence based on the continuous acquisition strategy of intelligent weighing sensor array and Internet of Things with a pre-set smoothing coefficient. The fluctuation threshold is used to perform fluctuation judgment analysis on the fluctuation index weighted moving sequence to detect repeated jump data and remove duplicates, so as to obtain the weighted data of individual poultry.

[0015] Experimental verification archives of poultry weight were obtained by big data network retrieval. Reference weight of individual poultry and the range of normal weight variation of individual poultry based on the reference weight were extracted from the experimental verification archives. Based on the range of normal weight variation, an effective feasible region of normal poultry weight variation and a preset barrier objective function for iterating weight data within the effective feasible region were established.

[0016] Taylor expansion algorithm is introduced. Based on the deduplicated weight data, the barrier objective function is calculated by Taylor expansion of the Hessian matrix in the Taylor expansion algorithm to obtain the local quadratic confidence radius of the deduplicated weight data iterating normal weight fluctuation in the effective feasible region.

[0017] Based on big data networks and poultry farming experience cases, the optimal curve statistical data of poultry growth patterns are extracted. Based on the optimal curve statistical data, a linear growth model of poultry age-weight is constructed. Based on the linear growth model of poultry age-weight, the age-weight trust constraint step size is preset.

[0018] Based on the age-weight trust constraint step size, the local quadratic trust radius corresponding to the deduplication weight data of each poultry individual is updated in the effective feasible region to obtain the normal weight approximate solution of the barrier objective function. If the normal weight approximate solution has reached the optimal state and cannot be updated further, the barrier term of the barrier objective function is excluded by self-consistent scaling, and the abnormal weight value of the poultry individual is output and filtered and bundled as invalid weight data.

[0019] Invalid weight data are cleaned by performing a cleaning process to obtain a preliminary set of weight data samples.

[0020] Preferably, the preset fluctuation threshold is calculated based on a smoothing coefficient preset by the intelligent weighing sensor array and the Internet of Things, which calculates the fluctuation index-weighted moving average sequence of real-time weight data in the processing time sequence. The fluctuation threshold is used to perform fluctuation judgment analysis on the fluctuation index-weighted moving average sequence to detect repeated jumps and remove duplicate data, thereby obtaining the de-weighted weight data of individual poultry. Specifically, this includes the following steps:

[0021] Obtain the sensing response index of the intelligent weighing sensor array and the observation acquisition delay of the Internet of Things, and preset a smoothing coefficient for continuous acquisition of weight data based on the sensing response index and the observation acquisition delay.

[0022] The processing time sequence of real-time weight data by the edge computing gateway is obtained. Based on the smoothing coefficient, the processing time sequence of real-time weight data is exponentially weighted and shifted. This makes the exponentially weighted shift highly sensitive to slight amplitude shifts when poultry individuals are weighed. A fluctuating exponentially weighted shift sequence of continuously collected real-time weight data is generated.

[0023] The continuous acquisition period of real-time weight data is defined as the fluctuation detection interval. The difference between adjacent real-time weight data points is compared within the fluctuation detection interval based on the fluctuation index-weighted moving sequence. Potential change point locations are enumerated and statistically analyzed, and the change point statistical results are output.

[0024] A preset fluctuation threshold is used to construct an upper and lower fluctuation sensitivity criterion for continuous sampling of poultry individuals with the processing time sequence, based on the fluctuation threshold, and to generate an upper fluctuation sensitivity limit and a lower fluctuation sensitivity limit.

[0025] If the weight fluctuation indicated in the weighted moving average of the volatility index is greater than the upper volatility sensitivity limit or less than the lower volatility sensitivity limit, it indicates that the slight movement of the poultry individual when stepping on or off the scale causes repeated fluctuations in the weight reading. In this case, the maximum statistic is extracted from the change point statistical results.

[0026] Within the fluctuation detection interval, the change point position corresponding to the maximum statistic is divided into multiple sub-change point intervals. The above fluctuation threshold is repeated, which is called the repeated fluctuation judgment step. Each sub-change point interval is recursively detected, and the change point position of the sub-change point interval is recorded during the recursive detection process and marked as a single change point position.

[0027] Redundancy is removed from the real-time weight data corresponding to the single variable point to obtain the weight data of individual poultry.

[0028] Preferably, step S103 specifically includes the following steps:

[0029] Construct a local density domain, and dynamically estimate the local window scale of the data sample density distribution within the local density domain based on the standard deviation and sample number corresponding to the initially screened weight data sample set. Adaptively preset the optimal kernel bandwidth for each weight data sample based on the local window scale.

[0030] The density clustering characteristic components that reflect the weight distribution of poultry are extracted from the initially screened weight data samples. The kernel density algorithm is introduced, and the local density contribution of all weight data samples is calculated radially symmetrically based on the density clustering characteristic components. The radially symmetric kernel function of each weight data sample is proposed.

[0031] Obtain the cluster neighborhood radius of the weight data sample set, anchor a certain weight data sample as the current drift reference data point, calculate the Mahalanobis distance between the current drift reference data point and the other weight data samples, and locate the density center of the poultry weight cluster distribution based on the Mahalanobis distance and the radial symmetric kernel function.

[0032] The kernel weights of all weight data samples within the cluster neighborhood radius are calculated based on the optimal kernel bandwidth. The weighted mean of each weight data sample is obtained. The weighted mean is then shifted to a weighted position along the upward direction that maintains the minimum distance interval from the density center to generate the mean drift vector of the weight data samples.

[0033] Repeat the above steps for calculating the mean shift vector to perform multiple local density gradient iterations on the regional clustering of the remaining weight data samples, and output the current migration norm of the mean shift vector.

[0034] If the current migration norm is less than the preset migration norm, it is considered that the sample point where the weight data sample is located has converged to the maximum local density peak. At this time, the iterative operation of multiple local density gradients is stopped, and the mean drift clustering result is generated.

[0035] Based on the mean-shift clustering results, outlier data points caused by equipment failure, environmental interference, or abnormal individuals are removed to obtain the effective weight dataset for poultry aggregation weighing.

[0036] Preferably, step S104 specifically includes the following steps:

[0037] The quality assessment index of poultry weight data distribution is obtained. Based on the quality assessment index, the data distribution feature dimension is preset. Based on the data distribution feature dimension, the basic statistics of the data distribution characteristics of each effective weight data in the effective weight dataset are calculated to obtain the distribution mean and distribution standard deviation of each effective weight data.

[0038] A low-dimensional embedding feature domain of the skewness-kurtosis plane is established, and the symmetry and sharpness of the effective weight data under each data distribution feature dimension are calculated to obtain the distribution skewness and distribution kurtosis of the effective weight data.

[0039] Based on the distribution skewness and kurtosis, the skewness-kurtosis plane coordinates of each effective weight data are concatenated, and a higher-order moment feature vector is constructed under the corresponding data distribution feature dimension for each skewness-kurtosis plane coordinate based on the distribution mean and distribution standard deviation;

[0040] The higher-order moment feature vectors are encoded and embedded into a low-dimensional embedding feature domain using sparse coding techniques. After encoding and embedding, the higher-order moment features are normalized and scaled, and then weighted and fused with each other with the original features of the effective weight data through mutual self-attention, thereby generating the skewness-kurtosis distribution plane of the effective weight dataset.

[0041] Obtain the distribution discrimination rules for poultry weight, set the symmetry threshold and sharpness threshold for poultry weight data to present an ideal normal distribution through the distribution discrimination rules, and locate and construct the normal distribution base point neighborhood on the skewness-kurtosis plane based on the symmetry threshold and sharpness threshold.

[0042] If the normal distribution base point covers the skewness-kurtosis distribution plane, then the effective weight dataset is considered and labeled to approximately follow a normal distribution, and enters the correction procedure for conventional mean calculation, outputting a statistical distribution characteristic analysis result.

[0043] If the normal distribution base point does not cover or intersect with the skewness-kurtosis distribution plane, then the effective weight dataset is considered and labeled to approximately follow a positively skewed distribution, and a corresponding specific correction mechanism program is initiated, outputting the results of the second type of statistical distribution characteristic analysis.

[0044] Preferably, step S105 specifically includes the following steps:

[0045] If the statistical distribution feature analysis results show that it is a type II statistical distribution feature analysis result, then the median absolute dispersion of the distribution shape of the effective weight dataset is evaluated and calculated by the standard deviation index method, and the transformation intensity of the effective weight dataset is preset according to the median absolute dispersion.

[0046] Based on the transform intensity, the generalized square root is used to determine the adaptive order of the square root transform for the positively skewed effective weight dataset. An empty stack of square root transform is constructed, and each effective weight data in the effective weight dataset is traversed.

[0047] During the traversal, an adaptive square root transformation is performed on each valid weight data based on the adaptive order, thereby compressing the weight of large values ​​in abnormal weight data. The transformed data is then imported into the square root transformation empty stack for storage, generating the valid weight data after square root transformation and the traversal reserve of the square root transformation empty stack.

[0048] If the traversal reserve amount reaches the preset traversal reserve amount, the effective weight dataset is marked as having undergone square root transformation, and a robust mean algorithm is introduced to calculate the effective weight data after square root transformation to obtain the robust transformation mean.

[0049] Obtain the original dimensional scale benchmark of the effective weight dataset before square root transformation, construct a dimensional inverse transformation criterion based on the original dimensional scale benchmark, and restore the mean of the robust transformation back to the original dimension through the dimensional inverse transformation criterion to obtain the original mean of the effective weight dataset.

[0050] The robust transformation mean is weighted and fused with the original mean to finally output the corrected result of the effective weight dataset. Based on the corrected result, the weight statistics and evenness evaluation value of poultry are determined.

[0051] A second aspect of the present invention provides a poultry weight cleaning and uniformity assessment system for breeding decision-making, applied to implement the poultry weight cleaning and uniformity assessment method for breeding decision-making as described in any one of the claims. The system specifically includes:

[0052] The weight data acquisition module includes an intelligent weighing sensor array and a wireless communication terminal. It is responsible for collecting the weight data of individual poultry in the target breeding area in real time and transmitting it to the next data processing layer, providing data support for weight index statistics and uniformity assessment.

[0053] The Internet of Things (IoT) cloud platform module includes an edge computing gateway and a multi-stage data cleaning engine, which are used for distributed streaming reception and processing of real-time weight data of poultry flocks, while performing a progressive deduplication and cleaning preliminary data screening procedure to remove outliers and invalid data.

[0054] The intelligent analysis module is responsible for analyzing the statistical distribution characteristics of skewness and kurtosis of the effective weight dataset of poultry aggregate weighing, determining whether the poultry weight data approximately follows a normal distribution, and providing an analytical basis for the correction of the weight data.

[0055] The intelligent correction module is used to perform square root transformation on the mean correction of the effective weight dataset of poultry aggregate weighing that exhibits a positively skewed distribution, so as to preserve the overall characteristics of the original weight data while eliminating the influence of outliers.

[0056] The data visualization interface is used to display accurate weight statistics and uniformity assessment values ​​of poultry weighing, as well as trend lines and visual charts of poultry growth dynamics.

[0057] The breeding decision support module is responsible for adjusting breeding decisions based on the visual content of accurate weight statistics and evenness assessment values, and providing a reserve of decisions for improving poultry breeding.

[0058] This invention addresses the technical deficiencies in the prior art, and its beneficial technical effects are as follows:

[0059] Based on the constraints of the poultry age-growth model and the trust planning of the reference weight ratio range, this method achieves accurate removal of invalid weighing data, effectively solving the problems of high misjudgment rate and poor adaptability caused by the reliance on fixed thresholds in traditional methods, thus significantly improving the quality of the initial data. Based on the mean-shift clustering algorithm, it automatically identifies the peak regions of weight data density, determines the main clusters through adaptive kernel bandwidth, and automatically removes outlier data points, solving the problem of hidden noise caused by random weighing, abnormal individuals, and interference from the weighing platform in poultry weighing. This makes the cleaned weight data more objectively reflect the true growth status of poultry. A biased... The degree-kurtosis low-dimensional embedding feature domain performs high-order moment low-dimensional embedding statistical distribution feature analysis on the effective weight data after deduplication and cleaning. This can quickly determine whether the weight distribution follows a normal distribution, thus providing a scientific basis for subsequent data correction. By performing square root transformation and robust mean correction on the effective weight data with binary distribution features, the impact of abnormally large individuals on the overall statistical value can be effectively reduced. This makes the final average weight and evenness evaluation closer to the actual growth level of poultry, avoiding the sensitivity of traditional average statistics to skewed distributions and ensuring the robustness and reliability of weight statistics and evenness assessment. In summary, this invention can provide accurate weight statistics, evenness indicators, and growth trend visualization for aquaculture management systems, enabling key decision support such as feeding adjustments, density management, and slaughter time prediction based on real data. It has high real-time performance, high reliability, and high adaptability, significantly improving management efficiency and economic benefits in digital aquaculture scenarios. Attached Figure Description

[0060] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art 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 embodiments can be obtained from these drawings without creative effort.

[0061] Figure 1 A first method flowchart for a poultry weight washing and uniformity assessment method for breeding decision-making is shown;

[0062] Figure 2 A second method flowchart for assessing poultry weight washing and uniformity for breeding decisions is shown.

[0063] Figure 3 A system framework diagram of a poultry weight cleaning and uniformity assessment system for breeding decision-making is shown. Detailed Implementation

[0064] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0065] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0066] The first aspect of this invention provides a method for assessing poultry weight washing and uniformity for breeding decision-making, such as... Figure 1 As shown, it includes the following steps:

[0067] S101: Collect real-time weight data of poultry individuals in the target breeding area through an intelligent weighing sensor array, detect duplicate data of real-time weight data by setting a fluctuation threshold and deduplicating the data, and after deduplication, filter invalid data of weight data based on the reference weight and corresponding proportion range of poultry individuals and the linear growth model of weight by age to obtain a preliminary set of weight data samples.

[0068] S102: If the number of samples in the initially screened weight data sample set is greater than the preset threshold, then proceed to the mean drift clustering calculation program.

[0069] S103: Upload the initially screened weight data sample set to the cloud platform. Based on the local density distribution and density clustering characteristics of the weight data sample set, preset the optimal kernel bandwidth and radial symmetric kernel function. Calculate the mean drift clustering of the weight dataset using the optimal kernel bandwidth and radial symmetric kernel function. Finally, remove outlier data points with weighing errors based on the mean drift clustering results to obtain the effective weight dataset of poultry cluster weighing.

[0070] S104: Establish a low-dimensional embedding feature domain for the skewness-kurtosis plane, calculate the distribution shape of the effective weight dataset, obtain the distribution skewness and distribution kurtosis, embed the higher-order moment feature vector of the effective weight data into the low-dimensional embedding feature domain based on the distribution skewness and distribution kurtosis, perform threshold judgment analysis of statistical distribution features, and output the analysis results of one type of statistical distribution features and the analysis results of two types of statistical distribution features.

[0071] S105: If the results are displayed as a type II statistical distribution characteristic analysis, the effective weight data is transformed by square root, and a robust mean correction calculation is performed on the transformed effective weight data. The output is the weight statistical index and uniformity assessment value of the poultry for accurate weighing, and a trend line of poultry growth dynamics is plotted to make adjustments to breeding decisions.

[0072] It should be noted that the multi-level data cleaning pipeline of this invention mainly consists of three cleaning stages. In the first stage, the basic validity filtering stage, obviously abnormal data is quickly screened out based on preset fluctuation threshold rules, reference weight ratio ranges, and age-weight linear growth constraints. This stage mainly handles outliers caused by equipment failure, environmental interference, or obvious errors. Through this stage, approximately 60-70% of obviously abnormal data is effectively identified and eliminated, greatly reducing the burden on subsequent cleaning stages. The second stage, cluster analysis and denoising, uses unsupervised learning methods to deeply explore the intrinsic structure of the weight data. This stage automatically identifies valid weight ranges through density distribution characteristics, making it particularly suitable for processing data that falls within the threshold range but still exhibits distribution anomalies. The third stage is the statistical distribution correction stage. Instead of simply accepting or rejecting data, the system optimizes the data through intelligent mathematical transformations. This correction is not arbitrary manual adjustment but is based on mathematical verification of square root transformation and robust mean recovery theory, ensuring that the corrected weight data more accurately reflects the actual weight status of poultry.

[0073] Preferably, step S101 specifically includes the following steps:

[0074] The target breeding area is identified, and the weight of individual poultry within the target breeding area is collected in real time through an intelligent weighing sensor array to obtain real-time weight data of different poultry individuals.

[0075] A pre-set fluctuation threshold is used to calculate the fluctuation index weighted moving sequence of real-time weight data in the processing time sequence based on the continuous acquisition strategy of intelligent weighing sensor array and Internet of Things with a pre-set smoothing coefficient. The fluctuation threshold is used to perform fluctuation judgment analysis on the fluctuation index weighted moving sequence to detect repeated jump data and remove duplicates, so as to obtain the weighted data of individual poultry.

[0076] Experimental verification archives of poultry weight were obtained by big data network retrieval. Reference weight of individual poultry and the range of normal weight variation of individual poultry based on the reference weight were extracted from the experimental verification archives. Based on the range of normal weight variation, an effective feasible region of normal poultry weight variation and a preset barrier objective function for iterating weight data within the effective feasible region were established.

[0077] Taylor expansion algorithm is introduced. Based on the deduplicated weight data, the barrier objective function is calculated by Taylor expansion of the Hessian matrix in the Taylor expansion algorithm to obtain the local quadratic confidence radius of the deduplicated weight data iterating normal weight fluctuation in the effective feasible region.

[0078] Based on big data networks and poultry farming experience cases, the optimal curve statistical data of poultry growth patterns are extracted. Based on the optimal curve statistical data, a linear growth model of poultry age-weight is constructed. Based on the linear growth model of poultry age-weight, the age-weight trust constraint step size is preset.

[0079] Based on the age-weight trust constraint step size, the local quadratic trust radius corresponding to the deduplication weight data of each poultry individual is updated in the effective feasible region to obtain the normal weight approximate solution of the barrier objective function. If the normal weight approximate solution has reached the optimal state and cannot be updated further, the barrier term of the barrier objective function is excluded by self-consistent scaling, and the abnormal weight value of the poultry individual is output and filtered and bundled as invalid weight data.

[0080] Invalid weight data are cleaned by performing a cleaning process to obtain a preliminary set of weight data samples.

[0081] It should be noted that due to the high activity level of poultry individuals during traditional weighing processes, mechanical errors in weighing equipment, and environmental interference, the final collected and displayed poultry weight data often contains a large amount of noise and outliers. Furthermore, traditional weight data preprocessing methods lack the ability to understand poultry growth mechanisms, thus failing to remove invalid weight data that deviates from their growth characteristics based on a deep understanding of poultry growth patterns. To address this, this method utilizes existing literature, experimental data, and breeding experience data from large datasets to obtain reference weights for poultry at different stages and under different conditions, such as the typical weight reference for poultry of a certain age, and the permissible range of fluctuation above and below the "normal" weight. It considers valid weight data to fluctuate between 35% and 165% of the reference weight. This range has been rigorously experimentally verified, encompassing the normal weight variation range of poultry while effectively excluding obvious outliers. Subsequently, based on this, an effective feasible region is created, which is the set of feasible values ​​for weight data within this range that are considered to be within normal fluctuation. Simultaneously, a barrier objective function is constructed within the effective feasible region to iterate the weight data. This function represents the penalty imposed on real-time weight data if it deviates from the effective feasible region, meaning the risk of it being considered abnormal increases. The effective feasible region and barrier function mechanism constrain the real-time weight data within a reasonable fluctuation range of typical weight, thereby quantifying the boundary between normal and abnormal. This ensures that the anomaly screening iteration of real-time weight data always stays within the feasible region to the maximum extent, guaranteeing that the screening of abnormal data components follows the proportional constraints of poultry reference weight. A second-order approximation of the barrier objective function at the current real-time weight data point is performed using the function's value, gradient, and Hessian matrix at that point, approximating the function change within the proportional range. Through this approximation, the calculated local quadratic confidence radius can track the feasible search interval for normal poultry weight variation at the current real-time observation point. If the real-time weight data continues to be controlled within the function change range of this confidence radius, it can be considered normal. Thus, a quantitative confidence region is planned, meaning that if the real-time weight continues to fall within this local quadratic confidence radius, it can be considered within the normal fluctuation range of the reference weight. If the values ​​exceed the limit, the data may contain potential anomalies.

[0082] It should be noted that this method also extracts the optimal or typical weight gain curves of poultry at different ages from a large amount of breeding experience data, thereby constructing a linear weight gain model for poultry growth based on age. For example, the model sets a lower limit of 11 grams for the baseline weight, with an increase of 5 grams per day; and an upper limit of 80 grams, with an increase of 70 grams per day. Simultaneously, the model also sets protective thresholds to ensure that the weight range does not expand indefinitely, with an upper limit not exceeding 5500 grams and a lower limit not lower than 625 grams. The design of this linear weight gain model based on age fully considers the slow growth stage in the early stages of poultry growth and the rapid development stage in the later stages. The age-weight confidence constraint step size, based on the model's preset value, represents the limit of reasonable weight gain or loss in poultry within a daily cycle, providing a physiologically reasonable growth constraint for the daily weight changes of individual poultry, forming a dynamic and normal trajectory. Then, based on the set age-weight trust constraint step size, the local quadratic trust radius corresponding to the real-time weight data of each poultry individual is updated within the current effective feasible domain (referencing the variation range). That is, as the weight changes with age-growth patterns, the trust radius of the real-time weight data may increase or decrease. The normal weight steady state is found through the iterative solution of the barrier objective function. As the data is updated, it gradually approaches an optimal state, i.e., the approximate solution of normal weight, which can then identify whether the weight data of a poultry individual is stable and normal or may enter an abnormal state. When the approximate solution has converged, the real-time weight of the current poultry individual is considered to have stabilized in its normal state. At this time, self-consistent scaling excludes or weakens the barrier term in the barrier function, which is equivalent to unlocking outliers and separating them. Thus, the output of abnormal data is false weight components caused by group activities, weighing errors, and environmental factors, effectively reducing the penalty weight and avoiding false screening or missed screening of anomalies. When the deviation intensifies and is still not stable, the penalty term continues to play a role in promoting identification. This method employs a dual-model threshold calculation strategy, based on a deep understanding of poultry growth patterns. It considers both the natural constraints of poultry age on weight range and the dynamic adjustment capability of reference weight, thereby accurately and efficiently screening out abnormal and invalid weight data. This improves the accuracy and reliability of weight monitoring in aquaculture management, providing highly reliable data support for subsequent accurate weight estimation and aquaculture decisions.

[0083] Preferably, the preset fluctuation threshold is calculated based on a smoothing coefficient preset by the intelligent weighing sensor array and the Internet of Things, which calculates the fluctuation index-weighted moving average sequence of real-time weight data in the processing time sequence. The fluctuation threshold is used to perform fluctuation judgment analysis on the fluctuation index-weighted moving average sequence to detect repeated jumps and remove duplicate data, thereby obtaining the de-weighted weight data of individual poultry. Specifically, this includes the following steps:

[0084] Obtain the sensing response index of the intelligent weighing sensor array and the observation acquisition delay of the Internet of Things, and preset a smoothing coefficient for continuous acquisition of weight data based on the sensing response index and the observation acquisition delay.

[0085] The processing time sequence of real-time weight data by the edge computing gateway is obtained. Based on the smoothing coefficient, the processing time sequence of real-time weight data is exponentially weighted and shifted. This makes the exponentially weighted shift highly sensitive to slight amplitude shifts when poultry individuals are weighed. A fluctuating exponentially weighted shift sequence of continuously collected real-time weight data is generated.

[0086] The continuous acquisition period of real-time weight data is defined as the fluctuation detection interval. The difference between adjacent real-time weight data points is compared within the fluctuation detection interval based on the fluctuation index-weighted moving sequence. Potential change point locations are enumerated and statistically analyzed, and the change point statistical results are output.

[0087] A preset fluctuation threshold is used to construct an upper and lower fluctuation sensitivity criterion for continuous sampling of poultry individuals with the processing time sequence, based on the fluctuation threshold, and to generate an upper fluctuation sensitivity limit and a lower fluctuation sensitivity limit.

[0088] If the weight fluctuation indicated in the weighted moving average of the volatility index is greater than the upper volatility sensitivity limit or less than the lower volatility sensitivity limit, it indicates that the slight movement of the poultry individual when stepping on or off the scale causes repeated fluctuations in the weight reading. In this case, the maximum statistic is extracted from the change point statistical results.

[0089] Within the fluctuation detection interval, the change point position corresponding to the maximum statistic is divided into multiple sub-change point intervals. The above fluctuation threshold is repeated, which is called the repeated fluctuation judgment step. Each sub-change point interval is recursively detected, and the change point position of the sub-change point interval is recorded during the recursive detection process and marked as a single change point position.

[0090] Redundancy is removed from the real-time weight data corresponding to the single variable point to obtain the weight data of individual poultry.

[0091] It should be noted that during the collection of poultry individual weight data, the frequent activity or stress responses of the poultry can cause slight fluctuations and duplications in the continuously collected weight data due to their minor movements on the weighing platform. Therefore, accurately identifying and removing duplicate weight components is crucial. To address this, this method sets a smoothing coefficient based on the sensing response index of the intelligent weighing sensor array and the observation and acquisition latency of the Internet of Things when continuously acquiring real-time weight data of individual poultry. This smoothing coefficient determines the sensitivity and focus of observation for duplicate weight data, ensuring that the smoothing operation during real-time weight data detection neither over-filters nor misses real fluctuation components, thus maintaining high vigilance in capturing duplicate data with slight amplitude shifts. Next, an exponentially weighted shift algorithm is applied to the real-time weight data according to the processing sequence. This makes the algorithm more sensitive to subtle changes in poultry weight during weighing, generating a fluctuation-weighted shift sequence that highlights continuous, slight fluctuations. This reduces the interference of random noise on trend judgment and effectively amplifies the cumulative effect of fluctuations under continuous small-amplitude shifts, allowing small-drift, repeatedly sensed data to gradually emerge. Compared to traditional mean and sliding window mean detection calculations, the fluctuation-weighted shift sequence can detect continuous, slight fluctuations in poultry weight more quickly. Subsequently, based on the fluctuation-weighted shift sequence, the differences between adjacent weight data points are compared to find potential jump locations, i.e., candidate change points, resulting in a list of change point statistics to describe the jump intensity of continuous data collection at each data location. Since the action of poultry weighing or unweighing causes brief spikes or drops in body weight readings, this list of change point statistics can capture the data locations where spikes or drops occur, preparing for anomaly localization in subsequent duplicate detection and deduplication operations.

[0092] It should be noted that this method establishes an intelligent recognition mechanism based on preset fluctuation thresholds. Specifically, preset fluctuation thresholds are used to establish upper and lower sensitive boundaries for judging the direction of fluctuation. Based on the time-series of continuously collected data fluctuations and processing, a dynamic boundary range is constructed to determine "how much exceeds to be considered upper fluctuation" (upper fluctuation limit) and "how much falls below to be considered lower fluctuation" (lower fluctuation limit). The warning mechanism of the fluctuation sensitivity limits ensures that when a slight deviation occurs during continuous data collection, the weighted shift of the fluctuation index can quickly cross the boundary, thereby accurately detecting whether the poultry's activities on or off the weighing platform exceed the fluctuation limits. Furthermore, different poultry weights and different noise levels at different times mean that without setting sensitivity limits, it would be impossible to determine whether the fluctuation is a genuine behavior. The upper fluctuation sensitivity limit corresponds to the positive fluctuation increase phenomenon of poultry suddenly applying force, stepping on, or frequently stomping on the weighing platform, while the lower fluctuation sensitivity limit corresponds to the negative fluctuation increase phenomenon of poultry leaving the weighing platform with both feet off or frequently stomping on the weighing platform. If duplicate fluctuation data points are detected within the fluctuation detection interval, the change point location corresponding to the maximum statistic is divided into multiple sub-change point intervals to further refine the detection task within the local range. This is because weighing or unweighing actions are usually not a single jump, but a series of jumps. Through recursive segmentation, all minute change points can be peeled off and identified layer by layer, so that even the slightest data fluctuations within each sub-interval can be captured and identified, preventing the omission of small but real change points. This method can add a deduplication mechanism to the weighing acquisition and preliminary screening of poultry weight data, thereby automatically identifying and retaining high-value and meaningful weight changes, filtering out duplicate data caused by slight movements of poultry, and significantly improving the efficiency and accuracy of weight data collection.

[0093] Preferably, S103, as Figure 2 As shown, the specific steps include:

[0094] Construct a local density domain, and dynamically estimate the local window scale of the data sample density distribution within the local density domain based on the standard deviation and sample number corresponding to the initially screened weight data sample set. Adaptively preset the optimal kernel bandwidth for each weight data sample based on the local window scale.

[0095] The density clustering characteristic components that reflect the weight distribution of poultry are extracted from the initially screened weight data samples. The kernel density algorithm is introduced, and the local density contribution of all weight data samples is calculated radially symmetrically based on the density clustering characteristic components. The radially symmetric kernel function of each weight data sample is proposed.

[0096] Obtain the cluster neighborhood radius of the weight data sample set, anchor a certain weight data sample as the current drift reference data point, calculate the Mahalanobis distance between the current drift reference data point and the other weight data samples, and locate the density center of the poultry weight cluster distribution based on the Mahalanobis distance and the radial symmetric kernel function.

[0097] The kernel weights of all weight data samples within the cluster neighborhood radius are calculated based on the optimal kernel bandwidth. The weighted mean of each weight data sample is obtained. The weighted mean is then shifted to a weighted position along the upward direction that maintains the minimum distance interval from the density center to generate the mean drift vector of the weight data samples.

[0098] Repeat the above steps for calculating the mean shift vector to perform multiple local density gradient iterations on the regional clustering of the remaining weight data samples, and output the current migration norm of the mean shift vector.

[0099] If the current migration norm is less than the preset migration norm, it is considered that the sample point where the weight data sample is located has converged to the maximum local density peak. At this time, the iterative operation of multiple local density gradients is stopped, and the mean drift clustering result is generated.

[0100] Based on the mean-shift clustering results, outlier data points caused by equipment failure, environmental interference, or abnormal individuals are removed to obtain the effective weight dataset for poultry aggregation weighing.

[0101] It should be noted that poultry weight distribution often exhibits obvious clustering characteristics, but traditional weight measurement methods cannot automatically detect natural clusters in the data. To address this, this method first adaptively estimates the local data density around each sample point based on the standard deviation and sample size of the data samples, determining the local window scale used for kernel density estimation. This allows the system to implement an adaptive optimal bandwidth selection strategy. The choice of bandwidth parameter determines the size of the search neighborhood; data points with high local density are allocated smaller bandwidths to highlight their dense structure, while data points with low density are allocated larger bandwidths to avoid the influence of sparse noise. This effectively improves the sensitivity and stability of kernel density estimation in multi-density poultry activity scenarios, solving the problems of uneven distribution of poultry weight data and different weight concentrations at different stages, laying a reliable foundation for subsequent peak search. Next, density clustering feature components reflecting the poultry weight structure, such as multi-peak structures or dense area features, are extracted from the dataset. The kernel density algorithm is used to calculate the corresponding radially symmetric kernel function for each data sample. This radially symmetric kernel function defines the weight decay of sample points at different distances, measuring the weight contribution weight of the weight data sample to the local density of surrounding sample points, ultimately forming a local density description of real poultry clusters with optimal bandwidth control. Each weight sample becomes a density emission source, exerting an observable influence on surrounding samples. This constructs a continuous and differentiable poultry weight density field, enabling the mean drift to exhibit gradient-ascending characteristics and explicitly characterizing the natural clustering trend of effective poultry weight and out-of-group weight. Then, a fixed sample point is used as the current drift baseline data point, serving as a reference for observing the drift behavior of other data sample points. The similarity of this point to other data is measured using Mahalanobis distance, and combined with a radially symmetric kernel function, the density center of each data sample is located more accurately. This more precise density center location avoids the misjudgment of skewed or long-tailed data by Mahalanobis distance, ensuring that the mean drift is not simply shifted to the neighbor's average, but rather shifted to the true density peak direction. This significantly improves the accuracy of natural clustering results of poultry weight, especially under noise interference from equipment malfunctions, environmental disturbances, or out-of-group weight data from abnormal individuals.

[0102] It should be noted that within the cluster neighborhood radius of each sample, the kernel weights of all sample points are calculated based on the kernel function and optimal bandwidth. These kernel weights effectively suppress weight data sample points far from the density center, making the drift path more stable and more consistent with the true density manifold. Furthermore, by moving towards the density center in the direction corresponding to the minimum distance interval, the weighted mean is migrated to its weighted position, achieving unsupervised clustering. This ensures that the clustering results of weight data are more natural, stable, and reliable without relying on traditional hyperparameters. The mean drift vector clarifies the direction in which weight data samples should move to reach the density peak. The current migration norm represents the distance the weighted mean moves to its weighted position, signifying a gradual approach to its corresponding density peak, forming a natural cluster structure. This method can automatically discover natural clusters in poultry weight data after deduplication and cleaning without any prior knowledge. It is particularly suitable for data processing of poultry weighing clusters. After clustering, the system selects the cluster containing the most weight data points as the effective dataset, and removes weight data in other clusters as outliers. This effectively identifies and eliminates erroneous data caused by equipment failure, environmental interference, or abnormal individuals, thereby obtaining a high-quality, reliable poultry weight dataset that truly reflects the weight structure of poultry clusters and provides a reliable data foundation for breeding decisions.

[0103] Preferably, step S104 specifically includes the following steps:

[0104] The quality assessment index of poultry weight data distribution is obtained. Based on the quality assessment index, the data distribution feature dimension is preset. Based on the data distribution feature dimension, the basic statistics of the data distribution characteristics of each effective weight data in the effective weight dataset are calculated to obtain the distribution mean and distribution standard deviation of each effective weight data.

[0105] A low-dimensional embedding feature domain of the skewness-kurtosis plane is established, and the symmetry and sharpness of the effective weight data under each data distribution feature dimension are calculated to obtain the distribution skewness and distribution kurtosis of the effective weight data.

[0106] Based on the distribution skewness and kurtosis, the skewness-kurtosis plane coordinates of each effective weight data are concatenated, and a higher-order moment feature vector is constructed under the corresponding data distribution feature dimension for each skewness-kurtosis plane coordinate based on the distribution mean and distribution standard deviation;

[0107] The higher-order moment feature vectors are encoded and embedded into a low-dimensional embedding feature domain using sparse coding techniques. After encoding and embedding, the higher-order moment features are normalized and scaled, and then weighted and fused with each other with the original features of the effective weight data through mutual self-attention, thereby generating the skewness-kurtosis distribution plane of the effective weight dataset.

[0108] Obtain the distribution discrimination rules for poultry weight, set the symmetry threshold and sharpness threshold for poultry weight data to present an ideal normal distribution through the distribution discrimination rules, and locate and construct the normal distribution base point neighborhood on the skewness-kurtosis plane based on the symmetry threshold and sharpness threshold.

[0109] If the normal distribution base point covers the skewness-kurtosis distribution plane, then the effective weight dataset is considered and labeled to approximately follow a normal distribution, and enters the correction procedure for conventional mean calculation, outputting a statistical distribution characteristic analysis result.

[0110] If the normal distribution base point does not cover or intersect with the skewness-kurtosis distribution plane, then the effective weight dataset is considered and labeled to approximately follow a positively skewed distribution, and a corresponding specific correction mechanism program is initiated, outputting the results of the second type of statistical distribution characteristic analysis.

[0111] It should be noted that the distribution mean characterizes the central tendency of the data distribution, while the distribution standard deviation reveals the dispersion of the data distribution. Through quality assessment of the data distribution and adaptive selection of feature dimensions, the system maintains high robustness when dealing with weight data with different distribution patterns. Subsequently, a two-dimensional feature space specifically for distribution pattern analysis is established, namely the low-dimensional embedded feature domain of the skewness-kurtosis plane. This low-dimensional embedded feature domain allows the data distribution pattern to become a learnable abstract feature, enhancing the perception of important structural features such as skewness, heavy tails, or leptokurtics in the weight data sample distribution, achieving lightweight and visualized distribution discrimination. It is worth mentioning that distribution skewness and kurtosis are core indicators for judging whether weight data approximates a normal distribution; among them, skewness is an important indicator for measuring the symmetry of the data distribution. In poultry weighing data, a positively skewed distribution (skewness greater than 0) usually means that there are a few abnormally large individuals, or that the weighing equipment has produced systematic biases under certain circumstances. A negatively skewed distribution may indicate the presence of underdeveloped individuals or systematic errors in the weight data collection process. By accurately calculating the skewness coefficient, the distribution pattern of the data can be accurately determined, providing a basis for subsequent correction processing. Kurtosis reflects the sharpness of the data distribution. High kurtosis indicates that the data is mainly concentrated around the mean, with a relatively concentrated distribution; low kurtosis indicates that the data distribution is relatively dispersed. Ideally, the weight data of healthy poultry should show a moderate kurtosis value, neither too concentrated nor too dispersed. Furthermore, this method concatenates the distribution skewness and distribution kurtosis into a plane coordinate point for location guidance, giving each weight data point a clear spatial traceability guide in the skewness-kurtosis plane. The mean and standard deviation are used to construct higher-order moment eigenvectors for each coordinate, including combinations of second, third, or fourth moments, as shown in the following example: This elevates the expression of data distribution characteristics from raw data to structured statistical features. Since higher-order moment features provide a more comprehensive expression of distribution patterns, they can more accurately predict distribution patterns and trends, thereby enhancing the ability to distinguish normal distributions.

[0112] It should be noted that by using sparse coding to effectively encode and embed higher-order moment features into a low-dimensional skewness-kurtosis feature domain, the higher-order moment information is transformed into low-dimensional, distribution-aware, but more discriminative embedded features. Furthermore, a self-attention mechanism is used to weight-fuse the higher-order moment features with the weight data features. This self-attention mechanism gives higher weight to important data distribution feature dimensions, creating a semantically enhanced response with the real weight data. Ultimately, a skewness-kurtosis distribution plane that comprehensively expresses the normal distribution characteristics of the weight data is obtained, achieving the effect of discriminative analysis of visualized statistical distribution features. The symmetry threshold under a normal distribution is generally 0.5, and the sharpness threshold is also generally 0.5. The underlying meaning of the normal distribution base point neighborhood covering the skewness-kurtosis distribution plane is that when the weight data simultaneously meets the conditions that the absolute value of skewness is less than 0.5 and the absolute value of kurtosis is less than 0.5, then the weight data is considered to approximately follow a normal distribution. This method can comprehensively analyze the distribution pattern of weight data from two dimensions: skewness and kurtosis. It can automatically identify the skewness characteristics of the data, provide a basis for subsequent error correction of weight data, and significantly improve the stability and decision reliability of weight monitoring.

[0113] Preferably, step S105 specifically includes the following steps:

[0114] If the statistical distribution feature analysis results show that it is a type II statistical distribution feature analysis result, then the median absolute dispersion of the distribution shape of the effective weight dataset is evaluated and calculated by the standard deviation index method, and the transformation intensity of the effective weight dataset is preset according to the median absolute dispersion.

[0115] Based on the transform intensity, the generalized square root is used to determine the adaptive order of the square root transform for the positively skewed effective weight dataset. An empty stack of square root transform is constructed, and each effective weight data in the effective weight dataset is traversed.

[0116] During the traversal, an adaptive square root transformation is performed on each valid weight data based on the adaptive order, thereby compressing the weight of large values ​​in abnormal weight data. The transformed data is then imported into the square root transformation empty stack for storage, generating the valid weight data after square root transformation and the traversal reserve of the square root transformation empty stack.

[0117] If the traversal reserve amount reaches the preset traversal reserve amount, the effective weight dataset is marked as having undergone square root transformation, and a robust mean algorithm is introduced to calculate the effective weight data after square root transformation to obtain the robust transformation mean.

[0118] Obtain the original dimensional scale benchmark of the effective weight dataset before square root transformation, construct a dimensional inverse transformation criterion based on the original dimensional scale benchmark, and restore the mean of the robust transformation back to the original dimension through the dimensional inverse transformation criterion to obtain the original mean of the effective weight dataset.

[0119] The robust transformation mean is weighted and fused with the original mean to finally output the corrected result of the effective weight dataset. Based on the corrected result, the weight statistics and evenness evaluation value of poultry are determined.

[0120] It should be noted that the median absolute dispersion is a quantitative description of the skewness of the structure, while the transform strength is the magnitude of the square root transform required, including strong and weak transform magnitudes. The stronger the data skewness, the greater the compression required; the weaker the skewness, the more minimal the root transform needed, preserving the effective structure of the weight data to the greatest extent. The adaptive order automatically calculates the exponential parameter of the square root transform based on the magnitude of the data skewness, ensuring that the transform strength corresponds to the degree of skewness in the weight data. This achieves flexible compression of heavy-tailed data, avoiding the "one-size-fits-all" distortion caused by a fixed square root transform, and making the square root transform correction of effective weight data adaptive and robust.

[0121] It should be noted that if the statistical distribution characteristic analysis results show a binary statistical distribution characteristic analysis result, it indicates that the effective weight dataset exhibits a positively skewed distribution. To address this, this method performs a square root transformation on the effective weight data, effectively compressing the weight of larger values ​​and reducing the impact of outliers on the overall mean. This makes the transformed data closer to a symmetrical and concentrated distribution (normal distribution). A robust mean of the transformed data is then calculated. A robust mean is a central location estimate that is insensitive to extreme values, avoiding the dominance of heavy-tailed samples in the statistical results and improving the representativeness and stability of the mean. Finally, the robust mean is inversely transformed back to its original dimensional scale based on the original dimensional scale of the effective weight data before the square root transformation, thereby restoring the original data scale. This allows the robust mean to form a mean estimate consistent with the original data dimensions. Based on skewness, a fusion weight is set to weight and fuse the robust transformed mean and the original mean. When the skewness is large, the transformed mean is more relied upon; when the skewness is small, more information from the original mean is retained, achieving a smooth transition. On the one hand, the overall characteristics of the original weight data are preserved, and on the other hand, the adverse effects of outliers are eliminated, ensuring that the corrected weight data more accurately reflects the actual weight status of the poultry population.

[0122] The second aspect of this invention provides a poultry weight washing and uniformity assessment system for breeding decision-making, such as... Figure 3 As shown, the system, applied to implement any of the poultry weight washing and uniformity assessment methods for breeding decision-making, specifically includes:

[0123] The weight data acquisition module includes an intelligent weighing sensor array and a wireless communication terminal. It is responsible for collecting the weight data of individual poultry in the target breeding area in real time and transmitting it to the next data processing layer, providing data support for weight index statistics and uniformity assessment.

[0124] The Internet of Things (IoT) cloud platform module includes an edge computing gateway and a multi-stage data cleaning engine, which are used for distributed streaming reception and processing of real-time weight data of poultry flocks, while performing a progressive deduplication and cleaning preliminary data screening procedure to remove outliers and invalid data.

[0125] The intelligent analysis module is responsible for analyzing the statistical distribution characteristics of skewness and kurtosis of the effective weight dataset of poultry aggregate weighing, determining whether the poultry weight data approximately follows a normal distribution, and providing an analytical basis for the correction of the weight data.

[0126] The intelligent correction module is used to perform square root transformation on the mean correction of the effective weight dataset of poultry aggregate weighing that exhibits a positively skewed distribution, so as to preserve the overall characteristics of the original weight data while eliminating the influence of outliers.

[0127] The data visualization interface is used to display accurate weight statistics and uniformity assessment values ​​of poultry weighing, as well as trend lines and visual charts of poultry growth dynamics.

[0128] The breeding decision support module is responsible for adjusting breeding decisions based on the visual content of accurate weight statistics and evenness assessment values, and providing a reserve of decisions for improving poultry breeding.

[0129] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A poultry weight cleaning and uniformity evaluation method for farming decision making, characterized by, Includes the following steps: S101: Collect real-time weight data of poultry individuals in the target breeding area through an intelligent weighing sensor array, detect duplicate data of real-time weight data by setting a fluctuation threshold and deduplicating the data, and after deduplication, filter invalid data of weight data based on the reference weight and corresponding proportion range of poultry individuals and the linear growth model of weight by age to obtain a preliminary set of weight data samples. S102: If the number of samples in the initially screened weight data sample set is greater than the preset threshold, then proceed to the mean drift clustering calculation program. S103: Upload the initially screened weight data sample set to the cloud platform. Based on the local density distribution and density clustering characteristics of the weight data sample set, preset the optimal kernel bandwidth and radial symmetric kernel function. Calculate the mean drift clustering of the weight dataset using the optimal kernel bandwidth and radial symmetric kernel function. Finally, remove outlier data points with weighing errors based on the mean drift clustering results to obtain the effective weight dataset of poultry cluster weighing. S104: Establish a low-dimensional embedding feature domain for the skewness-kurtosis plane, calculate the distribution shape of the effective weight dataset, obtain the distribution skewness and distribution kurtosis, embed the higher-order moment feature vector of the effective weight data into the low-dimensional embedding feature domain based on the distribution skewness and distribution kurtosis, perform threshold judgment analysis of statistical distribution features, and output the analysis results of one type of statistical distribution features and the analysis results of two types of statistical distribution features. S105: If the results are displayed as a type II statistical distribution characteristic analysis, the effective weight data is transformed by square root, and a robust mean correction calculation is performed on the transformed effective weight data. The output is the weight statistical index and uniformity assessment value of the poultry for accurate weighing, and a trend line of poultry growth dynamics is plotted to make adjustments to breeding decisions.

2. The poultry weight washing and uniformity assessment method for farming decision making as claimed in claim 1, wherein, S101 specifically includes the following steps: The target breeding area is identified, and the weight of individual poultry within the target breeding area is collected in real time through an intelligent weighing sensor array to obtain real-time weight data of different poultry individuals. A pre-set fluctuation threshold is used to calculate the fluctuation index weighted moving sequence of real-time weight data in the processing time sequence based on the continuous acquisition strategy of intelligent weighing sensor array and Internet of Things with a pre-set smoothing coefficient. The fluctuation threshold is used to perform fluctuation judgment analysis on the fluctuation index weighted moving sequence to detect repeated jump data and remove duplicates, so as to obtain the weighted data of individual poultry. Experimental verification archives of poultry weight were obtained by big data network retrieval. Reference weight of individual poultry and the range of normal weight variation of individual poultry based on the reference weight were extracted from the experimental verification archives. Based on the range of normal weight variation, an effective feasible region of normal poultry weight variation and a preset barrier objective function for iterating weight data within the effective feasible region were established. Taylor expansion algorithm is introduced. Based on the deduplicated weight data, the barrier objective function is calculated by Taylor expansion of the Hessian matrix in the Taylor expansion algorithm to obtain the local quadratic confidence radius of the deduplicated weight data iterating normal weight fluctuation in the effective feasible region. Based on big data networks and poultry farming experience cases, the optimal curve statistical data of poultry growth patterns are extracted. Based on the optimal curve statistical data, a linear growth model of poultry age-weight is constructed. Based on the linear growth model of poultry age-weight, the age-weight trust constraint step size is preset. Based on the age-weight trust constraint step size, the local quadratic trust radius corresponding to the deduplication weight data of each poultry individual is updated in the effective feasible region to obtain the normal weight approximate solution of the barrier objective function. If the normal weight approximate solution has reached the optimal state and cannot be updated further, the barrier term of the barrier objective function is excluded by self-consistent scaling, and the abnormal weight value of the poultry individual is output and filtered and bundled as invalid weight data. Invalid weight data are cleaned by performing a cleaning process to obtain a preliminary set of weight data samples.

3. The poultry weight washing and uniformity assessment method for farming decision making as claimed in claim 2, wherein, The preset fluctuation threshold is used to calculate the fluctuation index-weighted moving average sequence of real-time weight data in the processing time sequence based on the continuous acquisition strategy of the intelligent weighing sensor array and the Internet of Things with a preset smoothing coefficient. The fluctuation threshold is used to perform fluctuation judgment analysis on the fluctuation index-weighted moving average sequence to detect repeated jumps and remove duplicate data, so as to obtain the weight-free weight data of individual poultry. The specific steps include: Obtain the sensing response index of the intelligent weighing sensor array and the observation acquisition delay of the Internet of Things, and preset a smoothing coefficient for continuous acquisition of weight data based on the sensing response index and the observation acquisition delay. The processing time sequence of real-time weight data by the edge computing gateway is obtained. Based on the smoothing coefficient, the processing time sequence of real-time weight data is exponentially weighted and shifted. This makes the exponentially weighted shift highly sensitive to slight amplitude shifts when poultry individuals are weighed. A fluctuating exponentially weighted shift sequence of continuously collected real-time weight data is generated. The continuous acquisition period of real-time weight data is defined as the fluctuation detection interval. The difference between adjacent real-time weight data points is compared within the fluctuation detection interval based on the fluctuation index-weighted moving sequence. Potential change point locations are enumerated and statistically analyzed, and the change point statistical results are output. A preset fluctuation threshold is used to construct an upper and lower fluctuation sensitivity criterion for continuous sampling of poultry individuals with the processing time sequence, based on the fluctuation threshold, and to generate an upper fluctuation sensitivity limit and a lower fluctuation sensitivity limit. If the weight fluctuation indicated in the weighted moving average of the volatility index is greater than the upper volatility sensitivity limit or less than the lower volatility sensitivity limit, it indicates that the slight movement of the poultry individual when stepping on or off the scale causes repeated fluctuations in the weight reading. In this case, the maximum statistic is extracted from the change point statistical results. Within the fluctuation detection interval, the change point position corresponding to the maximum statistic is divided into multiple sub-change point intervals. The above fluctuation threshold is repeated, which is called the repeated fluctuation judgment step. Each sub-change point interval is recursively detected, and the change point position of the sub-change point interval is recorded during the recursive detection process and marked as a single change point position. Redundancy is removed from the real-time weight data corresponding to the single variable point to obtain the weight data of individual poultry.

4. The method for assessing poultry weight washing and uniformity based on breeding decision-making according to claim 1, characterized in that, S103 specifically includes the following steps: Construct a local density domain, and dynamically estimate the local window scale of the data sample density distribution within the local density domain based on the standard deviation and sample number corresponding to the initially screened weight data sample set. Adaptively preset the optimal kernel bandwidth for each weight data sample based on the local window scale. The density clustering characteristic components that reflect the weight distribution of poultry are extracted from the initially screened weight data samples. The kernel density algorithm is introduced, and the local density contribution of all weight data samples is calculated radially symmetrically based on the density clustering characteristic components. The radially symmetric kernel function of each weight data sample is proposed. Obtain the cluster neighborhood radius of the weight data sample set, anchor a certain weight data sample as the current drift reference data point, calculate the Mahalanobis distance between the current drift reference data point and the other weight data samples, and locate the density center of the poultry weight cluster distribution based on the Mahalanobis distance and the radial symmetric kernel function. The kernel weights of all weight data samples within the cluster neighborhood radius are calculated based on the optimal kernel bandwidth. The weighted mean of each weight data sample is obtained. The weighted mean is then shifted to a weighted position along the upward direction that maintains the minimum distance interval from the density center to generate the mean drift vector of the weight data samples. Repeat the above steps for calculating the mean shift vector to perform multiple local density gradient iterations on the regional clustering of the remaining weight data samples, and output the current migration norm of the mean shift vector. If the current migration norm is less than the preset migration norm, it is considered that the sample point where the weight data sample is located has converged to the maximum local density peak. At this time, the iterative operation of multiple local density gradients is stopped, and the mean drift clustering result is generated. Based on the mean-shift clustering results, outlier data points caused by equipment failure, environmental interference, or abnormal individuals are removed to obtain the effective weight dataset for poultry aggregation weighing.

5. The farming decision oriented poultry weight washing and uniformity evaluation method according to claim 1, characterized in that, S104 specifically includes the following steps: The quality assessment index of poultry weight data distribution is obtained. Based on the quality assessment index, the data distribution feature dimension is preset. Based on the data distribution feature dimension, the basic statistics of the data distribution characteristics of each effective weight data in the effective weight dataset are calculated to obtain the distribution mean and distribution standard deviation of each effective weight data. A low-dimensional embedding feature domain of the skewness-kurtosis plane is established, and the symmetry and sharpness of the effective weight data under each data distribution feature dimension are calculated to obtain the distribution skewness and distribution kurtosis of the effective weight data. Based on the distribution skewness and kurtosis, the skewness-kurtosis plane coordinates of each effective weight data are concatenated, and a higher-order moment feature vector is constructed under the corresponding data distribution feature dimension for each skewness-kurtosis plane coordinate based on the distribution mean and distribution standard deviation; The higher-order moment feature vectors are encoded and embedded into a low-dimensional embedding feature domain using sparse coding techniques. After encoding and embedding, the higher-order moment features are normalized and scaled, and then weighted and fused with each other with the original features of the effective weight data through mutual self-attention, thereby generating the skewness-kurtosis distribution plane of the effective weight dataset. Obtain the distribution discrimination rules for poultry weight, set the symmetry threshold and sharpness threshold for poultry weight data to present an ideal normal distribution through the distribution discrimination rules, and locate and construct the normal distribution base point neighborhood on the skewness-kurtosis plane based on the symmetry threshold and sharpness threshold. If the normal distribution base point covers the skewness-kurtosis distribution plane, then the effective weight dataset is considered and labeled to approximately follow a normal distribution, and enters the correction procedure for conventional mean calculation, outputting a statistical distribution characteristic analysis result. If the normal distribution base point does not cover or intersect with the skewness-kurtosis distribution plane, then the effective weight dataset is considered and labeled to approximately follow a positively skewed distribution, and a corresponding specific correction mechanism program is initiated, outputting the results of the second type of statistical distribution characteristic analysis.

6. The farming decision oriented poultry weight washing and uniformity evaluation method according to claim 1, characterized in that, S105 specifically includes the following steps: If the statistical distribution feature analysis results show that it is a type II statistical distribution feature analysis result, then the median absolute dispersion of the distribution shape of the effective weight dataset is evaluated and calculated by the standard deviation index method, and the transformation intensity of the effective weight dataset is preset according to the median absolute dispersion. Based on the transform intensity, the generalized square root is used to determine the adaptive order of the square root transform for the positively skewed effective weight dataset. An empty stack of square root transform is constructed, and each effective weight data in the effective weight dataset is traversed. During the traversal, an adaptive square root transformation is performed on each valid weight data based on the adaptive order, thereby compressing the weight of large values ​​in abnormal weight data. The transformed data is then imported into the square root transformation empty stack for storage, generating the valid weight data after square root transformation and the traversal reserve of the square root transformation empty stack. If the traversal reserve amount reaches the preset traversal reserve amount, the effective weight dataset is marked as having undergone square root transformation, and a robust mean algorithm is introduced to calculate the effective weight data after square root transformation to obtain the robust transformation mean. Obtain the original dimensional scale benchmark of the effective weight dataset before square root transformation, construct a dimensional inverse transformation criterion based on the original dimensional scale benchmark, and restore the mean of the robust transformation back to the original dimension through the dimensional inverse transformation criterion to obtain the original mean of the effective weight dataset. The robust transformation mean is weighted and fused with the original mean to finally output the corrected result of the effective weight dataset. Based on the corrected result, the weight statistics and evenness evaluation value of poultry are determined.

7. A poultry weight cleaning and uniformity evaluation system for farming decision making, characterized by, The system, applied to the poultry weight washing and uniformity assessment method for breeding decision-making as described in any one of claims 1-6, specifically includes: The weight data acquisition module includes an intelligent weighing sensor array and a wireless communication terminal. It is responsible for collecting the weight data of individual poultry in the target breeding area in real time and transmitting it to the next data processing layer, providing data support for weight index statistics and uniformity assessment. The Internet of Things (IoT) cloud platform module includes an edge computing gateway and a multi-stage data cleaning engine, which are used for distributed streaming reception and processing of real-time weight data of poultry flocks, while performing a progressive deduplication and cleaning preliminary data screening procedure to remove outliers and invalid data. The intelligent analysis module is responsible for analyzing the statistical distribution characteristics of skewness and kurtosis of the effective weight dataset of poultry aggregate weighing, determining whether the poultry weight data approximately follows a normal distribution, and providing an analytical basis for the correction of the weight data. The intelligent correction module is used to perform square root transformation on the mean correction of the effective weight dataset of poultry aggregate weighing that exhibits a positively skewed distribution, so as to preserve the overall characteristics of the original weight data while eliminating the influence of outliers. The data visualization interface is used to display accurate weight statistics and uniformity assessment values ​​of poultry weighing, as well as trend lines and visual charts of poultry growth dynamics. The breeding decision support module is responsible for adjusting breeding decisions based on the visual content of accurate weight statistics and evenness assessment values, and providing a reserve of decisions for improving poultry breeding.

Images