Intelligent management method and device for breeding farm based on reproduction cycle regulation

Abstract

The present application relates to a kind of intelligent management method and device of pig farm based on breeding cycle regulation, belong to pig breeding technical field, the present application is by constructing multimodal time series fusion model, and using olfactory feature and visual feature is trained to multimodal time series fusion model, utilize the multimodal time series fusion model output sow best insemination window prediction curve, to thereby based on sow best insemination window prediction curve statistics each sow estrus state, and the estrus state of each sow is updated, finally set the batch production plan of sow, according to the estrus state of updated sow and the batch production plan of sow relevant reproduction cycle regulation is generated.The present application is combined with nuclear principal component analysis, linear discriminant analysis and minimum redundancy maximum correlation criterion carries out feature extraction, effectively eliminates redundant information, constructs high discriminant estrus smell dynamic fingerprint spectrum and vulva morphological change data, significantly improves the accuracy and robustness of estrus state identification.

Description

Technical Field

[0001] This invention relates to the field of technology, and in particular to a method and device for intelligent management of pig farms based on reproductive cycle regulation. Background Technology

[0002] Traditional pig farm management relies on individual tracking, resulting in dispersed estrus, mating, and farrowing times for sows. This leads to complex herd management, low production efficiency, and difficulties in disease control. In recent years, batch production models have emerged, using biotechnology to regulate the sow reproductive cycle, allowing sows to be mated and farrowed in a planned, centralized manner, achieving "all-in, all-out" management and effectively improving biosecurity. However, current pig farming lacks real-time inventory monitoring mechanisms, making it difficult for managers to promptly grasp the inventory dynamics and pig status of each batch, impacting the timeliness of decision-making. Summary of the Invention

[0003] This invention overcomes the shortcomings of the prior art and provides a method and device for intelligent management of pig farms based on reproductive cycle regulation.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows: This invention provides an intelligent management method for pig farms based on reproductive cycle regulation, comprising the following steps: By deploying hardware in the breeding pig pen and using the deployed hardware to collect gas characteristics and visual image data, olfactory characteristics and visual characteristics are obtained by extracting features from the gas characteristics and visual image data. A multimodal temporal fusion model is constructed based on a long short-term memory network, and the olfactory and visual features are used to train the multimodal temporal fusion model. The multimodal temporal fusion model is then used to output the prediction curve of the optimal insemination window for breeding pigs. Based on the prediction curve of the optimal insemination window for the breeding pigs, the estrus status of each breeding pig is statistically analyzed and updated. Set up a batch production plan for breeding pigs, and generate relevant reproductive cycle regulation based on the updated estrus status of breeding pigs and the batch production plan for breeding pigs.

[0005] Furthermore, in the intelligent management method for breeding pig farms based on reproductive cycle regulation, hardware is deployed in the breeding pig pens, and the deployed hardware is used to collect gas characteristics and visual image data, specifically: A MEMS gas sensor array is deployed above the feeding trough in the sow gestation pen or large pen to collect the gas molecule components of the sow's exhaled gas and vulvar secretions. A depth camera is deployed directly above the pen to collect 3D point cloud data of the sow's back contour, ear base, and vulva area.

[0006] Furthermore, in the intelligent management method for breeding pig farms based on reproductive cycle regulation, olfactory and visual features are obtained by extracting features from the gas characteristics and visual image data, specifically: Kernel principal component analysis and linear discriminant analysis algorithms are introduced. Radial basis function kernels are used to map the original high-dimensional features of the gas to the regenerated kernel Hilbert space, where principal component extraction is performed. The covariance matrix is ​​processed by adding a Tikhonov regularization term to the inter-class scatter matrix of LDA using the linear discriminant analysis algorithm. The minimum redundancy and maximum relevance criterion is adopted. During the feature selection stage, features that are related to the estrus state but have redundancy levels below a preset redundancy threshold are retained first to obtain the processed feature vector. Based on the processed feature vectors, they are stacked in chronological order to form a time-feature matrix. The time-feature matrix is ​​visualized as a pseudo-color heatmap to form a dynamic fingerprint spectrum of estrus odor. The dynamic fingerprint spectrum of estrus odor is then used to identify real-time odor features and obtain olfactory features.

[0007] Furthermore, the intelligent management method for pig farms based on reproductive cycle regulation also includes: Using a pose estimation algorithm based on the Transformer architecture, we analyze the depth video stream in real time and extract the static pressure index and vulvar morphological changes. When extracting the static pressure index, when a boar or a sonic device stimulates the sow's back to bear virtual pressure, the change in spinal curvature is calculated. If the change in spinal curvature is stable within a predetermined threshold range, the current static pressure index is extracted. When extracting morphological changes of the vulva, an image segmentation algorithm is used to calculate the swelling degree and color saturation of the vulva region. The swelling degree and color saturation of the vulva region are used as quantitative indicators of the estrus process to form vulva morphological change data.

[0008] Furthermore, in the intelligent management method for breeding pig farms based on reproductive cycle regulation, a multimodal temporal fusion model is constructed based on a long short-term memory network. The olfactory and visual features are used to train the multimodal temporal fusion model, which then outputs a prediction curve for the optimal insemination window for the breeding pigs. Specifically: Typical estrus odor features are collected using dynamic fingerprinting of estrus odor, and the similarity between all olfactory features generated within a preset time period and typical estrus odor features is calculated. Olfactory features with similarity greater than a preset similarity threshold are assigned higher weights and aggregated into olfactory feature vectors. For continuous video streams, a visual feature vector is extracted once within a preset time period, then averaged using a sliding window and aligned with time points. A multimodal temporal fusion model is then constructed based on a long short-term memory network. The olfactory and visual feature vectors are input into the gated fusion unit to dynamically determine the contribution weight of each modality. The model outputs a continuous success rate curve, which represents the success probability of insemination at a given time point. A hybrid density network is used as the output layer. Monte Carlo Dropout estimation calculates the variance of the prediction curve by performing a preset number of forward propagations on the same input during inference. As new data is continuously collected, the model updates the prediction curve in real time.

[0009] Furthermore, in the intelligent management method for breeding pig farms based on reproductive cycle regulation, the multimodal time-series fusion model is used to output the prediction curve of the optimal insemination window for breeding pigs, specifically as follows: The olfactory and visual features of breeding pigs within a preset time period are obtained, and the olfactory and visual features of breeding pigs within the preset time period are input into the multimodal temporal fusion model for prediction. The optimal insemination window prediction curve for breeding pigs is obtained through prediction, and the optimal insemination window prediction curve for breeding pigs is then visualized.

[0010] Furthermore, the intelligent management method for pig farms based on reproductive cycle regulation also includes: Establish a complete batch system covering both production and non-production batches, where production batches include: mating batches, pregnancy batches, parturition batches, and lactation batches; Non-production batches include: piglet batches, nursery batches, growing batches, replacement batches, and quarantine batches. Each batch has a unique batch number and is configured with batch attributes, including: batch type, creation time, expected end time, number of pigs, pen location, and responsible person.

[0011] Furthermore, the intelligent management method for pig farms based on reproductive cycle regulation also includes: Immunization integration, health care integration, breeding integration, delivery integration, weaning integration, and standardization integration are carried out on a batch basis, and the time distribution and progress status of all batches are displayed in the form of Gantt charts or timelines. For each batch, basic information, inventory status, pig list, work records, production indicators, individual traceability function, and anomaly labeling are displayed.

[0012] A second aspect of the present invention provides an intelligent management device for pig farms based on reproductive cycle regulation, including a memory and a processor. The memory includes a program for an intelligent management method for pig farms based on reproductive cycle regulation. When the program for the intelligent management method for pig farms based on reproductive cycle regulation is executed by the processor, it implements the steps of the intelligent management method for pig farms based on reproductive cycle regulation as described in any one of the present invention.

[0013] This invention addresses the shortcomings of the prior art and has the following beneficial effects: This invention combines principal component analysis, linear discriminant analysis, and the minimum redundancy maximum correlation criterion for feature extraction, effectively eliminating redundant information and constructing a highly discriminative dynamic fingerprint of estrus odor and vulvar morphological change data, significantly improving the accuracy and robustness of estrus state identification. Furthermore, a multimodal temporal fusion model is constructed based on a long short-term memory network, dynamically adjusting the contribution weights of olfactory and visual features through gated fusion units, and using a hybrid density network to output a continuous success rate prediction curve. Monte Carlo Dropout estimation is combined to quantify prediction uncertainty, enabling the model to continuously update prediction results based on real-time data, providing accurate window period judgments for boar insemination and effectively improving insemination success rate and conception rate. Moreover, Markov chains are used to perform temporal modeling of the estrus state of each boar, achieving dynamic updating and prediction of estrus state by calculating state transition probabilities. This method overcomes the subjectivity and lag of traditional manual observation, accurately tracking state changes during the estrus process and providing reliable data support for subsequent reproductive regulation. Finally, by combining the updated estrus status with batch production plans, this invention enables the system to automatically determine whether the estrus of breeding pigs is concentrated within a preset time period and to automatically trigger estrus control measures for individuals that deviate from the plan. This method effectively improves the estrus synchronization rate of breeding pig herds, provides an intelligent management tool for intensive breeding pig farms to achieve all-in, all-out batch production, and significantly improves production efficiency and resource utilization. Attached Figure Description

[0014] 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.

[0015] Figure 1 A flowchart illustrating the overall process of intelligent management of pig farms based on reproductive cycle regulation is presented. Figure 2 A diagram of an intelligent management device for pig farms based on reproductive cycle regulation is shown. Detailed Implementation

[0016] 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.

[0017] 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.

[0018] like Figure 1 As shown, this invention provides an intelligent management method for pig farms based on reproductive cycle regulation, comprising the following steps: By deploying hardware in the breeding pig pen and using the deployed hardware to collect gas characteristics and visual image data, olfactory and visual characteristics are obtained by extracting features from the gas characteristics and visual image data. A multimodal temporal fusion model was constructed based on a long short-term memory network. The model was trained using olfactory and visual features. The model was then used to output a prediction curve for the optimal insemination window of breeding pigs. The estrus status of each breeding pig is statistically analyzed based on the prediction curve of the optimal insemination window, and the estrus status of each breeding pig is updated. Set up a batch production plan for breeding pigs, and generate relevant reproductive cycle regulation based on the updated estrus status of breeding pigs and the batch production plan for breeding pigs.

[0019] It should be noted that this invention forms a complete closed-loop management process, from hardware deployment, data collection, feature extraction, fusion modeling to status updates and production control. The entire process is highly automated, reducing manual intervention and management costs, while improving the scientific and precise level of breeding management, resulting in significant economic benefits and application value.

[0020] Furthermore, in the intelligent management method for breeding pig farms based on reproductive cycle regulation, hardware is deployed in the breeding pig pens, and the deployed hardware is used to collect gas characteristics and visual image data, specifically: A MEMS gas sensor array is deployed above the feed trough and in the tail area of ​​the sow's gestation pen or large pen to collect the gas molecule components of the sow's exhaled gas and vulvar secretions. A depth camera is deployed directly above the pen to collect 3D point cloud data of the sow's back contour, ear base, and vulva area.

[0021] Furthermore, in the intelligent management method for breeding pig farms based on reproductive cycle regulation, olfactory and visual features are obtained by extracting features from gas characteristics and visual image data, specifically: Kernel principal component analysis and linear discriminant analysis algorithms are introduced. Radial basis function kernels are used to map the original high-dimensional features of the gas to the regenerated kernel Hilbert space, where principal components are extracted. It should be noted that by using a radial basis function (RBF) kernel to map the original high-dimensional features to the regenerated kernel Hilbert space, principal component extraction is performed in this space, which can capture nonlinear relationships in odor features, such as synergistic / antagonistic effects between different compounds, and retain more discriminative information than linear PCA.

[0022] The covariance matrix is ​​processed by adding a Tikhonov regularization term to the inter-class scatter matrix of LDA using the linear discriminant analysis algorithm. The minimum redundancy and maximum relevance criterion is adopted. During the feature selection stage, features that are related to the estrus state but have redundancy levels below a preset redundancy threshold are retained first to obtain the processed feature vector. The processed feature vectors are stacked in chronological order to form a time-feature matrix. The time-feature matrix is ​​visualized as a pseudo-color heatmap to form a dynamic fingerprint of estrus odor. The dynamic fingerprint of estrus odor is then used to identify real-time odor features and obtain olfactory features.

[0023] It should be noted that the kernel method is introduced to solve the nonlinear problem of odor features, and the static fingerprint is upgraded to a dynamic fingerprint spectrum through temporal stacking, so that the system can accurately identify the evolution stage of the estrus cycle, rather than just a binary judgment of estrus / non-estrus.

[0024] Furthermore, the intelligent management method for pig farms based on reproductive cycle regulation also includes: Using a pose estimation algorithm based on the Transformer architecture, we analyze the depth video stream in real time and extract the static pressure index and vulvar morphological changes. When extracting the static pressure index, when a boar or a sound-simulating device is stimulated, the change in spinal curvature of the sow's back under virtual pressure is calculated. If the change in spinal curvature is stable within a predetermined threshold range, the current static pressure index is extracted. When extracting morphological changes of the vulva, an image segmentation algorithm is used to calculate the swelling degree and color saturation of the vulva region. The swelling degree and color saturation of the vulva region are used as quantitative indicators of the estrus process to form vulva morphological change data.

[0025] It should be noted that the swelling degree (i.e., area change rate) and color saturation (RGB to HSV space analysis) of the vulva region are calculated using image segmentation algorithms as quantitative indicators of the estrus process.

[0026] Furthermore, in the intelligent management method for breeding pig farms based on reproductive cycle regulation, a multimodal temporal fusion model is constructed based on a long short-term memory network. This model is trained using olfactory and visual features, and then uses the multimodal temporal fusion model to output a prediction curve for the optimal insemination window for breeding pigs. Specifically: Typical estrus odor features are collected using dynamic fingerprinting of estrus odor, and the similarity between all olfactory features generated within a preset time period and typical estrus odor features is calculated. Olfactory features with similarity greater than a preset similarity threshold are assigned higher weights and aggregated into olfactory feature vectors. For continuous video streams, a visual feature vector is extracted once within a preset time period, then averaged using a sliding window and aligned with time points. A multimodal temporal fusion model is then constructed based on a long short-term memory network. The olfactory and visual feature vectors are input into the gated fusion unit to dynamically determine the contribution weight of each modality. The model outputs a continuous success rate curve, which represents the success probability of insemination at a given time point. A hybrid density network is used as the output layer. Monte Carlo Dropout estimation calculates the variance of the prediction curve by performing a preset number of forward propagations on the same input during inference. As new data is continuously collected, the model updates the prediction curve in real time.

[0027] It should be noted that a multimodal temporal fusion model is constructed based on a long short-term memory network. The contribution weights of olfactory and visual features are dynamically adjusted through gated fusion units, and a hybrid density network is used to output a continuous success rate prediction curve. Combined with Monte Carlo Dropout estimation to quantify prediction uncertainty, the model can continuously update prediction results based on real-time collected data, providing accurate window period judgments for pig insemination and effectively improving insemination success rate and conception rate.

[0028] Furthermore, in the intelligent management method for breeding pig farms based on reproductive cycle regulation, a multimodal time-series fusion model is used to output the prediction curve of the optimal insemination window for breeding pigs, specifically: Acquire the olfactory and visual features of breeding pigs within a preset time period, and input the olfactory and visual features of breeding pigs within the preset time period into a multimodal temporal fusion model for prediction; The prediction curve of the optimal insemination window for breeding pigs is obtained through prediction, and the prediction curve of the optimal insemination window for breeding pigs is visualized.

[0029] Furthermore, the intelligent management method for pig farms based on reproductive cycle regulation also includes: Establish a complete batch system covering both production and non-production batches, where production batches include: mating batches, pregnancy batches, parturition batches, and lactation batches; Non-production batches include: piglet batches, nursery batches, growing batches, replacement batches, and quarantine batches. Each batch has a unique batch number and is configured with batch attributes, including: batch type, creation time, expected end time, number of pigs, pen location, and responsible person.

[0030] It should be noted that the mating batch includes records of mating time, mating method, and boar information; the gestation batch includes tracking gestation days and ultrasound results; the farrowing batch includes records of expected delivery date, actual farrowing time, and number of piglets; the lactation batch includes tracking lactation days and piglet survival rate; the piglet batch includes the piglet group from weaning to nursery; the nursery batch includes pigs in the nursery stage (usually 3-10 weeks after weaning); the growing batch includes pigs in the growing stage (usually 10 weeks to before slaughter); the replacement batch includes the selected replacement breeding pig group; and the quarantine batch includes newly introduced pigs or pigs under quarantine for disease observation.

[0031] Each batch has a unique batch number (format: batch type code + year + batch number, such as PS202401 representing the first breeding batch in 2024). Batch attributes include: batch type, creation time, expected end time, number of pigs, pen, responsible person, etc.

[0032] Furthermore, the intelligent management method for pig farms based on reproductive cycle regulation also includes: Immunization integration, health care integration, breeding integration, delivery integration, weaning integration, and standardization integration are carried out on a batch basis, and the time distribution and progress status of all batches are displayed in the form of Gantt charts or timelines. For each batch, basic information, inventory status, pig list, work records, production indicators, individual traceability function, and anomaly labeling are displayed.

[0033] It should be noted that: Immunization integration includes automatically generating immunization plans based on batch age, supporting unified batch immunization operations, and recording vaccine type, dosage, immunization time, and executor; Health care integration includes developing batch health care plans (deworming, disinfection, drug addition, etc.), implementing and recording them in the batch; Mating integration includes centralized mating of sows within a batch, recording mating details, and automatically calculating the expected farrowing date; Farrowing integration includes centralized farrowing management in the batch, recording farrowing details, and automatically generating piglet batches; Weaning integration includes unified weaning in the batch, automatically transferring piglets to piglet batches, and sows entering the next mating batch; Standardized operations include establishing a batch standard operating procedure (SOP) library, including feeding standards, environmental control standards, health inspection standards, etc., and the system automatically pushes operation tasks.

[0034] It should be noted that the batch overview dashboard includes a Gantt chart or timeline display showing the time distribution and progress status of all batches, supporting filtering by batch type; basic information includes batch number, type, creation time, current age, and expected end time; inventory dynamics include current inventory, cumulative mortality, and inventory rate curve; the pig list includes the electronic ear tag number, sex, weight, and health status of all pigs in the batch; the work record includes a list of completed and pending immunization, health care, and breeding operations; production indicators include key performance indicators (KPIs) such as feed conversion ratio, daily weight gain, conception rate, and litter size; the individual traceability function allows clicking on any pig in the batch dashboard to view the complete life cycle information of that pig (historical batches, health records, pedigree information, etc.); and anomaly marking includes color-coded indicators of abnormal status in the batch dashboard (e.g., red indicates disease outbreak, yellow indicates low inventory rate, and green indicates normal).

[0035] Furthermore, in the intelligent management method for breeding pig farms based on reproductive cycle regulation, the estrus status of each breeding pig is statistically analyzed based on the prediction curve of the optimal insemination window, and the estrus status of each breeding pig is updated, specifically as follows: Set a threshold range for estrus status, and statistically determine the estrus status of each breeding pig within a preset time period from the pig's optimal insemination window prediction curve based on the threshold range for estrus status. For example, when a certain estrus state is within a certain threshold range, it is one of the following: proestrus, estrus, metestrus, and estrus. There are at least four threshold ranges, and each threshold range represents a estrus state.

[0036] A Markov chain is introduced to construct an estrus state matrix for each breeding pig based on the estrus state of each breeding pig within a preset time period. Input the estrus state matrix of the breeding pigs into a Markov chain and calculate the state transition probability value of the transition from one estrus state to another at each timestamp. Set a state transition probability threshold. When there is an estrus state node whose state transition probability value is greater than the state transition probability threshold, update the corresponding estrus state node to another estrus state node.

[0037] It should be noted that estrus states include proestrus, estrus, metestrus, and estrus. Since estrus states can transition from one state to another, such as from proestrus to estrus, Markov chains are used to perform time-series modeling of the estrus state for each breeding pig. By calculating the state transition probability, dynamic updates and predictions of estrus states are achieved, improving the accuracy of estrus state prediction. This method overcomes the subjectivity and lag of traditional manual observation, accurately tracking state changes during the estrus process and providing reliable data support for subsequent reproductive regulation.

[0038] Furthermore, in the intelligent management method for breeding pig farms based on reproductive cycle regulation, a batch production plan for breeding pigs is set up. Based on the updated estrus status of the breeding pigs and the batch production plan, relevant reproductive cycle regulation is generated, specifically as follows: Set up a batch production plan for breeding pigs, obtain the estrus status of each breeding pig within a preset time period based on the updated estrus status of the breeding pigs, and set the time period for the concentrated estrus of the breeding pigs; Determine whether the estrus status of each breeding pig within the preset time period is concentrated within the period of concentrated estrus of breeding pigs; When the estrus state of each breeding pig is concentrated within the period of concentrated estrus of breeding pigs within the preset time, there is no need to regulate the reproductive cycle. If the estrus state of each breeding pig does not coincide with the period when breeding pigs are in estrus, the remaining breeding pigs that are not in estrus during the period when breeding pigs are in estrus will be regulated.

[0039] It should be noted that this method, combined with the updated estrus status and batch production plan, allows the system to automatically determine whether the estrus of breeding pigs is concentrated within a preset time period, and automatically trigger estrus control measures for individuals deviating from the plan, such as drug control or physical control methods. This method effectively improves the estrus synchronization rate of breeding pig herds, provides intelligent management tools for intensive breeding pig farms to achieve all-in, all-out batch production, and significantly improves production efficiency and resource utilization.

[0040] In addition, this method also includes: A dynamically updated digital twin model is established for each breeding pig, and genotype information, historical reproductive data, and current hormone levels are integrated into the digital twin model; Based on the principles of population pharmacokinetics, a drug-organism interaction model is constructed. For key drugs in batch production, such as aceprogesterone, the absorption rate constant, clearance rate, and volume of distribution in the body of breeding pigs are calculated based on the current digital twin data of breeding pigs. Based on the absorption rate constant, clearance rate and distribution volume of the breeding pigs, a personalized drug intervention schedule for the breeding pigs is generated, including the time of first administration, the single administration dose and the injection dose of PMSG for each pig. During drug administration, the digital twin model is corrected in real time through visual and olfactory monitoring feedback. If a certain pig responds to the drug slower than the model predicts, the drug administration period is automatically extended or the subsequent PMSG dose is fine-tuned.

[0041] It should be noted that this method combines clinical pharmacology with digital twin technology, enabling a shift from group-based batch management to individualized, precise batch control, ensuring that the population achieves highly synchronized and high-quality ovulation within the predetermined time window.

[0042] It should be noted that the genotype information includes gene markers related to high litter size, historical reproductive data includes parity, weaning-estrus interval, and litter size, real-time body condition data includes feed intake and weight changes from automated feeding stations, and hormone levels are periodically detected using non-invasive saliva / tear biosensors.

[0043] In addition, this method also includes: An infrared thermal imager is deployed under the farrowing crate to monitor the temperature field distribution of the sow's udder. By analyzing the changes in the entropy of the temperature field, the farrowing situation of the breeding pig is monitored, and the farrowing movements of the sow and the interval between piglets are monitored visually. If the sow is detected to be continuously straining for more than a preset time without producing piglets, or if the interval between piglet births exceeds a preset time, a level three dystocia alarm will be triggered immediately. Physical stimulation will be performed by gently tapping the sow's ribs or lower abdomen with a robotic arm to assist in delivery, and veterinary personnel will be notified at the same time. When the sow changes from standing to lying down, her lying speed and the position of the piglets are analyzed in real time. The sow's lying trajectory is predicted by Kalman filter. If the predicted trajectory coincides with the piglets, intervention is immediately carried out by directional ultrasound or directional strong sound wave to drive the piglets away from the danger zone. By visually recognizing the umbilical cord status and activity level of newborn piglets, a robotic arm is guided to assist weaker piglets in finding teats and suckling colostrum.

[0044] It should be noted that this method integrates multimodal precision delivery early warning and autonomously intervenes in the two core causes of death during sow farrowing: dystocia and piglet crushing, which greatly reduces the pressure of manual delivery assistance.

[0045] like Figure 2As shown, the second aspect of the present invention provides an intelligent management device for pig farms based on reproductive cycle regulation, including a memory and a processor. The memory includes a program for an intelligent management method for pig farms based on reproductive cycle regulation. When the program for the intelligent management method for pig farms based on reproductive cycle regulation is executed by the processor, it implements any of the steps of the intelligent management method for pig farms based on reproductive cycle regulation.

[0046] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0047] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0048] In addition, in the various embodiments of the present invention, each functional unit can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0049] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0050] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

[0051] 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 method for intelligent management of pig farms based on reproductive cycle regulation, characterized in that, Includes the following steps: By deploying hardware in the breeding pig pen and using the deployed hardware to collect gas characteristics and visual image data, olfactory characteristics and visual characteristics are obtained by extracting features from the gas characteristics and visual image data. A multimodal temporal fusion model is constructed based on a long short-term memory network, and the olfactory and visual features are used to train the multimodal temporal fusion model. The multimodal temporal fusion model is then used to output the prediction curve of the optimal insemination window for breeding pigs. Based on the prediction curve of the optimal insemination window for the breeding pigs, the estrus status of each breeding pig is statistically analyzed and updated. Set up a batch production plan for breeding pigs, and generate relevant reproductive cycle regulation based on the updated estrus status of breeding pigs and the batch production plan for breeding pigs.

2. The intelligent management method for breeding pig farms based on reproductive cycle regulation according to claim 1, characterized in that, Hardware was deployed in the breeding pig pens, and the deployed hardware was used to collect gas characteristics and visual image data, specifically: A MEMS gas sensor array is deployed above the feeding trough in the sow gestation pen or large pen to collect the gas molecule components of the sow's exhaled gas and vulvar secretions. A depth camera is deployed directly above the pen to collect 3D point cloud data of the sow's back contour, ear base, and vulva area.

3. The intelligent management method for breeding pig farms based on reproductive cycle regulation according to claim 1, characterized in that, By extracting features from the gas characteristics and visual image data, olfactory and visual features are obtained, specifically: Kernel principal component analysis and linear discriminant analysis algorithms are introduced. Radial basis function kernels are used to map the original high-dimensional features of the gas characteristics to the regenerated kernel Hilbert space, and principal component extraction is performed in this space. The covariance matrix is ​​processed by adding a Tikhonov regularization term to the inter-class scatter matrix of LDA using the linear discriminant analysis algorithm. The minimum redundancy and maximum relevance criterion is adopted. During the feature selection stage, features that are related to the estrus state but have redundancy levels below a preset redundancy threshold are retained first to obtain the processed feature vector. Based on the processed feature vectors, they are stacked in chronological order to form a time-feature matrix. The time-feature matrix is ​​visualized as a pseudo-color heatmap to form a dynamic fingerprint spectrum of estrus odor. The dynamic fingerprint spectrum of estrus odor is then used to identify real-time odor features and obtain olfactory features.

4. The intelligent management method for breeding pig farms based on reproductive cycle regulation according to claim 3, characterized in that, Also includes: Using a pose estimation algorithm based on the Transformer architecture, we analyze the depth video stream in real time and extract the static pressure index and vulvar morphological changes. When extracting the static pressure index, when a boar or a sonic device stimulates the sow's back to bear virtual pressure, the change in spinal curvature is calculated. If the change in spinal curvature is stable within a predetermined threshold range, the current static pressure index is extracted. When extracting morphological changes of the vulva, an image segmentation algorithm is used to calculate the swelling degree and color saturation of the vulva region. The swelling degree and color saturation of the vulva region are used as quantitative indicators of the estrus process to form vulva morphological change data.

5. The intelligent management method for breeding pig farms based on reproductive cycle regulation according to claim 1, characterized in that, A multimodal temporal fusion model is constructed based on a long short-term memory network. The model is trained using olfactory and visual features. The model then outputs a prediction curve for the optimal insemination window for breeding pigs. Specifically: Typical estrus odor features are collected using dynamic fingerprinting of estrus odor, and the similarity between all olfactory features generated within a preset time period and typical estrus odor features is calculated. Olfactory features with similarity greater than a preset similarity threshold are assigned higher weights and aggregated into olfactory feature vectors. For continuous video streams, a visual feature vector is extracted once within a preset time period, then averaged using a sliding window and aligned with time points. A multimodal temporal fusion model is then constructed based on a long short-term memory network. The olfactory and visual feature vectors are input into the gated fusion unit to dynamically determine the contribution weight of each modality. The model outputs a continuous success rate curve, which represents the success probability of insemination at a given time point. A hybrid density network is used as the output layer. Monte Carlo Dropout estimation calculates the variance of the prediction curve by performing a preset number of forward propagations on the same input during inference. As new data is continuously collected, the model updates the prediction curve in real time.

6. The intelligent management method for breeding pig farms based on reproductive cycle regulation according to claim 1, characterized in that, The multimodal time-series fusion model is used to output the prediction curve for the optimal insemination window of breeding pigs, specifically as follows: The olfactory and visual features of breeding pigs within a preset time period are obtained, and the olfactory and visual features of breeding pigs within the preset time period are input into the multimodal temporal fusion model for prediction. The optimal insemination window prediction curve for breeding pigs is obtained through prediction, and the optimal insemination window prediction curve for breeding pigs is then visualized.

7. The intelligent management method for pig farms based on reproductive cycle regulation according to claim 1, characterized in that, Also includes: Establish a complete batch system covering both production and non-production batches, where production batches include: mating batches, pregnancy batches, parturition batches, and lactation batches; Non-production batches include: piglet batches, nursery batches, growing batches, replacement batches, and quarantine batches. Each batch has a unique batch number and is configured with batch attributes, including: batch type, creation time, expected end time, number of pigs, pen location, and responsible person.

8. The intelligent management method for breeding pig farms based on reproductive cycle regulation according to claim 7, characterized in that, Also includes: Immunization integration, health care integration, breeding integration, delivery integration, weaning integration, and standardization integration are carried out on a batch basis, and the time distribution and progress status of all batches are displayed in the form of Gantt charts or timelines. For each batch, basic information, inventory status, pig list, work records, production indicators, individual traceability function, and anomaly labeling are displayed.

9. A smart management device for pig farms based on reproductive cycle regulation, characterized in that, The system includes a memory and a processor. The memory includes a program for an intelligent management method for breeding pig farms based on reproductive cycle regulation. When the processor executes the program for the intelligent management method for breeding pig farms based on reproductive cycle regulation, it implements the steps of the intelligent management method for breeding pig farms based on reproductive cycle regulation as described in any one of claims 1-8.

Images