CN-122025122-A - Multi-mode perception-based intelligent evaluation system and method for health state of laying hens

CN122025122ACN 122025122 ACN122025122 ACN 122025122ACN-122025122-A

Abstract

The invention discloses an intelligent evaluation system and method for health status of laying hens based on multi-modal sensing, and relates to the technical field of intelligent breeding and animal health monitoring, wherein the system comprises a multi-modal data integration acquisition module, a dynamic multi-modal self-adaptive fusion module, a health-egg laying bidirectional collaborative prediction module, a closed loop iteration optimization module and a health-egg laying linkage suggestion output module; the invention realizes accurate and stable state evaluation by constructing a comprehensive and accurate data base, carrying out multidimensional acquisition, differential pretreatment and dynamic fusion, mining the data cooperative value, adopts the association of healthy-egg laying bidirectional cooperative prediction mining time and space, ensures the result to be attached to reality through closed loop optimization, outputs linkage suggestions, and defines a healthy regulation and control and egg laying optimization scheme, thereby providing scientific decision basis and power assisting refined management for breeding personnel.

Inventors

Hao Haiyu
WANG WENXIN
GE XIANGPING
Tan Xuejin
YANG PEIPEI
LIU YAWEN
WANG JINLONG
Lou Lanqiang
Wei Lianggong
LV CHANGJIANG
CHEN YONG
YANG LI

Assignees

青岛市畜牧工作站(青岛市畜牧兽医研究所)

Dates

Publication Date: 20260512
Application Date: 20260112

Claims (10)

1. The intelligent evaluation system for the health state of the laying hen based on multi-modal sensing is characterized by comprising a multi-modal data integration acquisition module, a dynamic multi-modal self-adaptive fusion module, a health-egg laying bidirectional collaborative prediction module, a closed-loop iteration optimization module and a health-egg laying linkage suggestion output module; The multi-mode data integration acquisition module acquires physiological, behavioral, fecal characteristics, environment, feed nutrition and breed age data of the laying hens, and forms a high-quality data set through pretreatment; The dynamic multi-mode self-adaptive fusion module is used for receiving a high-quality data set, and combining an improved attention algorithm to fuse the high-quality data set into multi-mode fusion data according to a judging technology of a health risk level and an egg laying prediction error; the health-egg laying bidirectional collaborative prediction module is used for receiving multi-mode fusion data, adopting a space-time sequence transducer model to mine data space-time correlation characteristics and outputting two core results of health risk level and abnormal cause of the laying hens, egg laying rate and egg quality change trend; The closed loop iteration optimization module is used for collecting monitoring data of the health risk and the laying rate of the laying hens, comparing the monitoring data with the output result of the health-egg laying bidirectional collaborative prediction module in a deviation way, and adjusting the fusion coefficient of the dynamic multi-mode self-adaptive fusion module according to the comparison result in a reverse iteration way to generate an optimized fusion coefficient; The health-egg laying linkage suggestion output module receives two types of core results of the health-egg laying bidirectional collaborative prediction module and an optimized fusion coefficient of the closed loop iteration optimization module, generates targeted linkage suggestions by combining a laying hen breeding rule, and forms system closed loop optimization through data real-time transmission and feedback.
2. The intelligent evaluation system based on multimodal perception of the health state of the laying hens is characterized in that the process of acquiring physiological, behavioral, fecal characteristics, environment, feed nutrition and breed day age data of the laying hens in the multimodal data integration acquisition module comprises the steps of sampling according to the proportion of 1000:1 of the total number of the laying hens in physiological data acquisition, fixing non-invasive acquisition equipment on the legs or the backs of the laying hens, recording the body temperature and heart rate of the laying hens, regularly taking blood samples to analyze cortisol levels, acquiring behavioral data through high-definition cameras uniformly arranged on the tops of a breeding house, recording the microscopic action changes of conventional activity pecking, drinking water and feeding of the laying hens, capturing fecal characteristic data through image acquisition equipment above a fecal channel, capturing odor information through a gas detection device, regularly taking fecal samples, analyzing the condition of a flora through a detection chip, acquiring environment data through acquisition devices uniformly arranged at ventilation openings and feeding areas in the breeding house, recording information of the temperature and humidity and illumination intensity, synchronizing time recording, extracting feed samples through detection equipment matched with the feeding devices, simultaneously recording the content of protein and energy content, recording the total amount and breed content, recording the total number and the breed content and the corresponding to the daily age of the breeding personnel and the corresponding to the number of the breeding personnel and the birth date and the system.
3. The intelligent evaluation system for the health state of the laying hens based on multi-modal sensing is characterized in that a high-quality data set is formed by preprocessing in the multi-modal data integration acquisition module, and specifically comprises the steps of screening various acquired original data, supplementing data which belong to random deletion according to corresponding data of other laying hens in the same period aiming at data deletion conditions, supplementing data which belong to continuous deletion in short time periods by a linear interpolation method, independently marking the data which belong to continuous deletion in long time periods, carrying out dimension unified processing on all data subjected to deletion value processing, wherein physiological data and environmental data are scaled by a normalization method, behavior data and fecal characteristic data are adjusted by a standardized method, all data are converted into the same numerical value interval, adjusting the time sequence of other modal data by a resampling technology based on the acquisition time sequence of the environmental data, enabling all modal data to correspond to each other in a time stamp, and finally integrating to form the high-quality data set.
4. The intelligent evaluation system for health status of laying hens based on multi-modal sensing according to claim 1, wherein the determination technology of the health risk level and the egg laying prediction error in the dynamic multi-modal adaptive fusion module is specifically as follows: The health risk level judgment comprises the steps of counting the number of indexes exceeding a preset normal threshold range and the exceeding amplitude of the indexes according to physiological indexes and fecal characteristic indexes in the multi-mode data, scoring by combining with quantitative abnormal indexes in behavior data, judging low health risk when no indexes exceed a threshold value and no abnormal behaviors exist, judging medium health risk when the number of the super-threshold indexes does not exceed 30% of the total number or slight quantitative abnormality exists, and judging high health risk when the number of the super-threshold indexes exceeds 30% of the total number or serious quantitative abnormality exists; The egg laying prediction error is judged by counting the average relative error according to the relative error of the daily egg laying prediction value, judging that the egg laying prediction error is smaller when the average relative error is smaller than 5%, judging that the egg laying prediction error is medium when the average relative error is between 5 and 10%, and judging that the egg laying prediction error is larger when the average relative error is larger than 10%.
5. The intelligent evaluation system for health status of laying hens based on multi-modal sensing according to claim 1, wherein the dynamic multi-modal adaptive fusion module combines an improved attention algorithm to fuse into multi-modal fusion data, and the formula is: , wherein, The multi-mode fusion data is finally output; Is the total number of modes participating in fusion; fusion weights for the ith modality; normalized feature vectors for the ith modality; the basic weight of the ith mode; correlating the weight for the health risk of the ith modality; correlating weights for egg laying errors of the ith mode; is the sum of the basic weight, the health risk association weight and the egg laying error association weight of all modes.
6. The intelligent evaluation system for health status of laying hens based on multi-modal sensing according to claim 1, wherein the health-egg laying bidirectional collaborative prediction module adopts a space-time sequence transducer model to mine data space-time correlation characteristics, and the formula is as follows: , wherein, The output result of the model; Respectively a query vector, a key vector and a value vector; Is a normalization function; Performing dot product operation for the transpose of the query vector Q and the key vector K; adjusting coefficients for the model; Dynamic weights of all modes; Is a scaling factor.
7. The intelligent evaluation system for the health state of the laying hen based on multi-modal sensing is characterized in that the process of outputting two core results of the health risk level and abnormal cause, the laying rate and the egg quality change trend of the laying hen in the health-egg laying bidirectional collaborative prediction module comprises the steps of receiving multi-modal fusion data transmitted by a dynamic multi-modal adaptive fusion module, inputting the multi-modal fusion data into a space-time sequence transducer model, mining space-time correlation characteristics in the data by the model, combining a health task branch of the model with a judging technology of a health risk level and egg production prediction error, simultaneously correlating a corresponding abnormal cause library to obtain an abnormal cause, predicting the laying rate according to the space-time correlation characteristics by the egg production task branch of the model, analyzing the egg quality correlation characteristics to obtain two core results of the health risk level and the abnormal cause, the laying rate and the egg quality change trend of the laying hen.
8. The intelligent evaluation system for health status of layers based on multi-modal perception according to claim 1 is characterized in that the specific process of deviation comparison and fusion coefficient adjustment in the closed loop iterative optimization module is that the deviation rate of health risk grades is calculated firstly, the deviation rate is obtained by dividing the number of layers with errors by the total number of layers to be monitored and multiplying by 100%, the prediction relative error of the egg rate is calculated again, the error is obtained by dividing the absolute difference between the prediction egg rate and the actual egg rate by the actual egg rate and multiplying by 100%, the total deviation is calculated later, the weight of the health risk grade deviation rate is 0.4, the weight of the prediction relative error of the egg rate is 0.6, the total deviation is equal to the sum of the health risk grade deviation rate multiplied by 0.4 and the prediction relative error of the egg rate multiplied by 0.6, the total deviation threshold is set to be 10%, when the total deviation is smaller than or equal to 10%, the current fusion coefficient of the dynamic multi-modal adaptive fusion module is unchanged, when the total deviation is larger than 10% and smaller than or equal to 20%, the fusion coefficient corresponding to each mode is reduced by the amplitude of 5% -10%, the fusion coefficient corresponding to each mode is reduced by the corresponding to the amplitude of the total deviation is reduced by the total deviation of the corresponding to the 20%, and the fusion coefficient is adjusted until the total deviation is reduced by the total deviation is smaller than 15%.
9. The intelligent evaluation system for the health state of the laying hen based on multi-modal sensing is characterized in that the step of generating targeted linkage advice by combining a laying hen breeding rule in the health-egg production linkage advice output module comprises the steps of firstly extracting health risk level, abnormal cause, egg yield and egg quality change trend output by the health-egg production bidirectional collaborative prediction module when advice is generated, then combining an optimized fusion coefficient of the closed loop iteration optimization module to construct an advice generation rule base, wherein the rule base comprises health regulation and control schemes and egg production optimization strategies under different scenes, determining regulation and control directions according to abnormal cause when specific advice is generated, adjusting regulation and control parameters by combining egg production performance prediction trends, verifying after advice generation, predicting health risk level change and egg production performance change after implementation advice according to multi-modal fusion data and optimized fusion coefficient, and finally outputting linkage advice in the form of a text report and a visual chart, wherein the text report clearly regulates and controls targets, specific operation steps and expected effects, and the visual chart shows the egg quality prediction trend and simulation comparison after regulation and control.
10. The intelligent evaluation method for the health state of the laying hens based on the multi-modal sensing is suitable for the intelligent evaluation system for the health state of the laying hens based on the multi-modal sensing, which is characterized by comprising the following specific steps of: s100, multi-mode data acquisition preprocessing, namely acquiring physiological, behavioral, fecal characteristics, environment, feed nutrition and variety day-to-day data of the laying hens, and carrying out deletion value complementation, dimension unification and time sequence alignment processing on the original data to form a high-quality data set; S200, multi-mode data self-adaptive fusion, namely determining data weight influence factors through a health risk level and egg laying prediction error judging technology, calculating each mode fusion weight by combining an improved attention algorithm, and fusing a high-quality data set to generate multi-mode fusion data; s300, health-egg laying bidirectional collaborative prediction, namely inputting multi-mode fusion data into a space-time sequence transducer model, mining data space-time correlation characteristics, obtaining health risk level and abnormal inducement through model health task branches, and obtaining egg laying rate and egg quality change trend through egg laying task branches; S400, closed loop iterative optimization of fusion coefficients, namely collecting actual monitoring data of health risks and laying rates of the laying hens, calculating a health risk level deviation rate, a laying rate prediction relative error and a comprehensive deviation, adjusting the fusion coefficients according to the comprehensive deviation amplitude, and iteratively optimizing until the comprehensive deviation reaches the standard to generate optimized fusion coefficients; s500, generating and outputting a linkage suggestion, namely extracting a bidirectional prediction core result and an optimization fusion coefficient, constructing a rule base comprising a health regulation scheme and an egg laying optimization strategy, generating and verifying a targeted linkage suggestion, and outputting the targeted linkage suggestion in a text report and visual chart form.

Description

Multi-mode perception-based intelligent evaluation system and method for health state of laying hens Technical Field The invention relates to the technical field of intelligent breeding and animal health monitoring, in particular to an intelligent evaluation system and an intelligent evaluation method for health states of laying hens based on multi-modal perception. Background The laying hen breeding is an important component part of livestock breeding industry, and the health state is directly related to the laying rate and the egg quality, so that the economic benefit of the breeding is decisive. Along with the popularization of large-scale and intensive culture modes, the culture density of the laying hens is improved, the influence of factors such as environment and nutrition on the health of the laying hens is more complex, and the traditional evaluation mode relying on manual observation and single-point monitoring is difficult to meet the real-time and accurate requirements. An evaluation system capable of integrating multidimensional data and realizing intelligent analysis is urgently needed, scientific support is provided for health regulation and control and egg laying optimization in the breeding process, and digital and intelligent transformation of laying hen breeding is promoted. The prior related technologies of egg laying health assessment and egg laying prediction have many defects that the data acquisition dimension is single, the data acquisition dimension is focused on environmental or physiological single-mode data, the synergistic effect of key factors such as behaviors, fecal characteristics, feed nutrition and the like is ignored, so that an assessment result is unilateral, the data fusion mode is fixed, the dynamic change of health risk level and egg laying prediction error is not considered, the weight of each mode data cannot be adaptively adjusted according to actual scenes, the effectiveness of fusion data is influenced, the health assessment and egg laying prediction are mutually independent, the space-time correlation characteristics between the health assessment and egg laying prediction are not mined, the bidirectional synergistic analysis is difficult to realize, the prediction precision is difficult to stabilize for a long time, the output advice is mostly a single-dimension regulation scheme, the targeted strategy of health and egg laying linkage is not formed, and the practicality and the suitability are insufficient. In conclusion, the prior art has obvious defects in the aspects of comprehensive data integration, adaptability of fusion modes, prediction cooperativity, suggestion pertinence and the like, and cannot meet the requirements of large-scale laying hen cultivation on accurate evaluation of health states and efficient regulation and control of laying eggs. Therefore, a set of intelligent evaluation system based on multi-mode sensing and with dynamic fusion, bidirectional collaborative prediction and closed-loop optimization functions is developed, linkage analysis and scientific suggestion output of health and egg laying are realized, and the intelligent evaluation system has important practical significance for solving the current pain point of laying hen breeding and improving the intelligent level of breeding. Disclosure of Invention The invention aims to make up the defects of the prior art and provides an intelligent evaluation system and method for the health state of a laying hen based on multi-mode perception, which can comprehensively acquire multi-dimensional data of the laying hen through a multi-mode data integration acquisition module and preprocess the multi-dimensional data; the system comprises a dynamic multi-mode self-adaptive fusion module, a health-egg laying bidirectional collaborative prediction module, a closed-loop iteration optimization module, a health-egg laying linkage suggestion output module, a system closed loop generation module and a power-assisted laying hen accurate cultivation module, wherein the dynamic multi-mode self-adaptive fusion module fuses data according to health risks and egg laying errors, the health-egg laying bidirectional collaborative prediction module utilizes model mining data characteristics to output health and egg laying core results, the closed-loop iteration optimization module adjusts fusion coefficients through deviation ratios, and the health-egg laying linkage suggestion output module generates suggestions according to cultivation rules. On one hand, the system comprises a multi-mode data integration and acquisition module, a dynamic multi-mode self-adaptive fusion module, a health-egg laying bidirectional collaborative prediction module, a closed loop iteration optimization module and a health-egg laying linkage suggestion output module; The multi-mode data integration acquisition module acquires physiological, behavioral, fecal characteristics, environment, feed nutrition and breed age data of the