Feeding method and device capable of accurately regulating and controlling feed intake of sows

Abstract

The embodiment of the invention discloses a feeding method and device capable of accurately regulating and controlling the feed intake of sows. The method comprises the following steps: identifying an individual identity through an electronic ear tag, and collecting backfat thickness, body condition score and weight data by using an inspection robot; dynamically calling a standard feeding curve based on the fetus number and the body condition score, and finely adjusting the target feed intake in combination with the body weight change trend; a quantitative or interactive feeding mode is adopted in the gestation period, feeding is carried out with different meals of wet mixed feed in the lactation period, and the feed-water ratio is dynamically adjusted; meanwhile, a data closed loop is constructed, a feeding strategy is optimized through big data analysis, and abnormal individuals are visualized. According to the scheme, precise self-adaptive regulation and control of the feeding behavior of the sow in the whole breeding cycle are achieved, and the feed efficiency and the breeding performance are remarkably improved.

Description

TECHNICAL FIELD

[0001] Embodiments of the present disclosure relate to the field of livestock breeding automation, in particular to a feeding method and device for precisely regulating the feed intake of sows. BACKGROUND

[0002] The number of breeding sows in China exceeds 40 million, which can provide nearly 700 million marketable pigs per year. The number of pigs in stock, the number of pigs marketed, and the output of pork all exceed half of the world. Although the pig farming scale is large, our breeding efficiency is relatively low, and the average number of piglets per sow per year (PSY) is only about 20. The per capita efficiency, production cost, and pig farming industry of the world's leading countries such as Denmark, the Netherlands, and Germany have a large gap. Comprehensive analysis of the reasons is that the quality of breeding pigs is low, the breeding process is backward, the level of facility equipment is low, and other factors, resulting in fewer piglets per sow, high mortality of piglets, low per capita production efficiency, serious waste of resources, and high feeding cost.

[0003] The sow feeding of a large-scale farm is usually divided into two stages: the empty pregnancy stage and the delivery and lactation stage. In the empty pregnancy stage, the corresponding feed nutrients need to be accurately provided according to the breed, parity, body condition, back fat, etc. of the sow, so as to ensure that the sow passes the gestation period smoothly and maintains good physical condition, so as to successfully deliver. In the delivery and lactation stage, the sow's feed intake needs to be limited in the early stage to avoid difficult delivery when delivering piglets, and the feeding amount needs to be gradually increased after the piglets are born to ensure that the sow can intake enough nutrients and provide sufficient milk for the piglets, so as to ensure the survival rate and growth rate of the piglets, and to reduce the back fat loss of the sow after weaning and maintain good body condition to quickly enter the next breeding cycle.

[0004] At present, in the empty pregnancy stage, most farms use a scale cup to measure the feed, and adjust the amount of feed for each sow through artificial recognition and judgment. The accuracy of the scale cup is very low, and factors such as installation height, installation angle, observation angle, and feed variety will cause measurement errors. This error will affect the sow's physical condition due to overfeeding or underfeeding. At the same time, the artificial recognition and judgment of the sow's demand for feed intake requires a high level of experience for the feeders, which is not conducive to standardized operation and difficult to improve efficiency. Therefore, frequent manual adjustment not only wastes manpower, but also has no efficiency, and is also prone to feed waste.

[0005] In the lactation stage of childbirth, most farms also use the method of feeding with a measuring cup. The amount of feeding is judged by artificial. Due to the difference in the feeding demand and the time period of the sow, the artificial feeding with a measuring cup has low precision, low efficiency, increases the cost of labor, and easily causes feeding waste. Some farms configure a simple mechanical free-feeding feeder, which allows the lactating sow to freely feed by touching the feeding control device. The mechanical feeder cannot control the amount of feeding, and it is easy to cause waste problems during the sow playing. And the feed left in the trough is easy to deteriorate, which can cause the sow and piglets to get sick and cause greater losses.

[0006] The current market precision feeding products usually use a preset standard feeding curve, rely on artificial judgment of the feeding demand of each sow to make individual adjustment, and increase or decrease the amount of feeding on the feeder. This method also needs experienced feeders to make comprehensive judgments on each sow, and the personal subjective factors are more, which is difficult to realize standardized operation. And due to the needs of pig farm production, sows need to circulate, and the feeder is usually fixed on the stall and does not move, so the feeder and the sow need to be re-bound and set up by artificial every time the sow circulates, which produces a large amount of work and difficulty. SUMMARY

[0007] The summary of the present disclosure is used to introduce the concepts in a brief form, which will be described in detail in the specific embodiments part. The summary of the present disclosure is not intended to identify the key features or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0008] Some embodiments of the present disclosure propose a feeding method and device for accurately regulating the feeding amount of sows. The method and device construct a dynamic feedback regulation mechanism with individual physiological state as the core variable, integrate multi-source heterogeneous data acquisition, feeding curve self-adaptive calling and fine-tuning, interactive feeding control logic, and full-process data closed-loop management, and fundamentally solve the structural contradiction between the static threshold control and the dynamic physiological demand of sows in the prior art.

[0009] In a first aspect, some embodiments of the present disclosure provide a feeding method for precisely regulating the feed intake of sows, comprising the following steps: first, collecting individual data of the sows, the individual data at least including breed, parity, body condition score, backfat thickness, gestational age or postpartum age; second, dynamically calling a corresponding feeding curve from a pre-stored standard feeding curve library based on the individual data; third, fine-tuning the feeding curve according to real-time body weight change data of the sows to generate individualized feeding parameters; fourth, based on the individualized feeding parameters, controlling the feeding equipment to perform quantitative or interactive unloading operations, and collecting real-time feed intake and water intake data; and finally, performing big data analysis on the collected data, optimizing the feeding strategy, and outputting a feeding state prompt.

[0010] In a second aspect, some embodiments of the present disclosure provide a sow intelligent feeding device for implementing the above-mentioned method, comprising:

[0011] a data collection module for collecting individual data of the sows;

[0012] a curve management module for storing a standard feeding curve library and dynamically calling and fine-tuning a feeding curve based on the individual data;

[0013] a unloading control module for controlling the feeding equipment to perform unloading operations according to the individualized feeding parameters;

[0014] a data analysis module for analyzing the feed intake and water intake data and outputting an optimized strategy and a feeding state.

[0015] The above-mentioned various embodiments of the present disclosure have the following beneficial effects:

[0016] Through the feeding method for precisely regulating the feed intake of sows according to some embodiments of the present disclosure, a complete technical system integrating individual recognition, multi-modal physiological parameter sensing, dynamic feeding curve generation, intelligent unloading execution and closed-loop data optimization is constructed, realizing precise, adaptive and full-process automatic regulation and control of the feed intake behavior of sows. This system discards the limitations of traditional static threshold control, regards sows as living beings with dynamic feedback capability, and its feeding strategy evolves in real time with the individual state, thereby maximizing feed conversion efficiency and reproductive performance while ensuring animal welfare. It realizes truly individualized precise feeding, and significantly improves breeding efficiency.

[0017] By employing inspection robots to automatically collect data on sows' backfat thickness, body condition scores, and weight changes, and combining this data with individual identification information (such as parity and breed) from electronic ear tags, the system can dynamically call upon and fine-tune pre-stored standard feeding curves to generate personalized feeding parameters highly matched to each sow's real-time physiological state. This data-driven dynamic adjustment mechanism overcomes the traditional extensive model that relies on fixed curves and manual experience, accurately meeting the nutritional needs of sows at different stages of non-pregnant pregnancy and lactation. This significantly reduces feed waste in pregnant sows while significantly increasing feed intake and milk production in lactating sows, thereby improving piglet survival rate and average weaning weight, and effectively reducing postpartum backfat loss in sows, shortening the interval before they enter the next reproductive cycle.

[0018] This invention significantly improves the automation and intelligence of feeding management, reducing reliance on manual experience. By integrating IoT technology and a big data analytics platform, it constructs a closed-loop intelligent management system. The system automatically executes the entire process from individual identification, data collection, curve matching, feed control to data feedback, eliminating the need for frequent manual intervention and judgment by farm workers. The management platform provides various data views, such as lists, groups, and pen cards, along with their filtering and sorting functions, enabling managers to quickly and intuitively locate sows with abnormal feeding behavior and perform batch parameter modifications. This greatly reduces the workload of farm workers, lowers reliance on individual experience, facilitates standardized and replicable farming management, and improves overall per capita management efficiency.

[0019] This invention provides a flexible, adaptable feeding mode with a waste prevention mechanism. The feed control system supports both quantitative and interactive feeding modes, and can intelligently switch according to the sow's stage (e.g., non-pregnant or farrowing / lactating). Especially during the farrowing / lactating stage, the system effectively avoids feed waste caused by excessive feeding by setting a maximum single feed amount and combining feedback from a feed residue detection sensor. Simultaneously, it supports setting the feed-to-water ratio based on age and allows for setting dry feed and wet feed (diluted feed) for different meal times. This improves feed palatability, promotes feed intake, and effectively prevents feed spoilage in the trough by avoiding nighttime feeding of wet feed, thus ensuring the health of the pig herd.

[0020] A continuously optimized data foundation has been established, providing strong support for breeding decisions. By collecting and uploading data such as feed intake, water consumption, and weight of each sow to the cloud server in real time, the system has accumulated valuable big data resources. Using this data for historical trend analysis, group comparison, and performance evaluation, preset feeding curves and strategies can be continuously verified and optimized, forming a continuous improvement closed loop of "data collection-analysis-optimization-execution." This transforms pig farm feeding management from static experience to dynamic data-driven approaches, continuously reducing long-term overall feeding costs and providing a scientific basis for production decisions.

[0021] In summary, this invention deeply integrates data collection, intelligent algorithms, the Internet of Things, and big data analysis to form a highly efficient, precise, and automated feeding management solution, which significantly improves the overall benefits of sow breeding from multiple dimensions such as precision nutrition, efficiency improvement, cost control, and animal welfare. Attached Figure Description

[0022] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0023] Figure 1 This is a schematic diagram of the overall architecture of a feeding method for precisely controlling feed intake in sows according to the present invention;

[0024] Figure 2 This is a schematic diagram of the data organization structure of the standard feeding curve library of this invention;

[0025] Figure 3 This is a schematic diagram of the feeding curve during the non-pregnant pregnancy feeding stage of the present invention. Detailed Implementation

[0026] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0027] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0028] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0029] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0030] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0031] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0032] The intelligent sow feeding system described in this invention consists of five functional units: a data acquisition module, a curve management module, a feed control module, a data analysis module, and a human-computer interaction module. These modules work collaboratively to form a complete technological closed loop. Figure 1 As shown in the diagram. The data acquisition module is responsible for obtaining individual sow identification information and multiple key physiological parameters; the curve management module dynamically calls and fine-tunes feeding strategies based on the acquired data; the feed control module executes quantitative or interactive feed instructions; the data analysis module processes, visualizes, and optimizes historical and real-time data; and the human-computer interaction module provides a user interface and decision support for farmers.

[0033] During the data collection phase, each sow wears an electronic ear tag that conforms to the ISO 11784 / 85 international standard. This tag has a built-in non-volatile storage unit for persistently storing the sow's unique identification code and basic profile information. This basic profile information includes breed code (e.g., LY, DL, Duroc), parity number (P1, P2, P3, Ps representing first, second, third, and multiple litters respectively), initial body condition score (BCS initial value), and mating or farrowing date. The electronic ear tag is read by a radio frequency identification antenna mounted on the inspection robot.

[0034] The inspection robot periodically travels along a preset path to each sow's positioning pen. It can integrate visible light cameras, 3D depth cameras, RFID identification modules, millimeter-wave radar ranging modules, infrared thermal imaging arrays, and / or structured light 3D scanners. When the robot stops in front of a positioning pen, for example, it can read the identification code from the electronic ear tag via an RFID antenna, and then simultaneously initiate multiple sensing and measurement tasks. The visible light camera acquires images of the sow's body surface at a rate of thirty frames per second, and identifies body condition features through edge computing algorithms. The 3D depth camera emits near-infrared structured light to acquire point cloud data of the sow's back contour, with a point cloud density of one hundred sampling points per square centimeter. The backfat estimation model is based on depth image feature extraction and multivariate regression analysis, and its calculation formula is as follows:

[0035]

[0036] Where BFT is backfat thickness, di is depth value, d is average depth, W is weight estimate, and αβ and γ are model coefficients. Backfat thickness is in millimeters, depth value is the raw distance data collected by the 3D camera, average depth is the arithmetic mean of the sampling area, weight estimate is calculated through a body size parameter regression model, and model coefficients are determined through training with 3,000 sets of sample data.

[0037] In a preferred embodiment, backfat thickness is estimated using a geometric model based on 3D point cloud data. The three-dimensional morphology of the sow's back is correlated with its backfat thickness; a thicker fat layer results in a fuller back profile and a larger radius of curvature in specific areas. Backfat thickness (BF) is estimated using the following empirical formula:

[0038]

[0039] Wherein, BFT: Backfat Thickness, in millimeters, refers to the estimated thickness of the fat layer between the skin and muscle layers at a specific point on the back of a sow. ρ: Point Cloud Density, a dimensionless ratio, refers to the number of valid data points contained in a 3D point cloud per unit projected area. This value reflects the richness of the data acquired by the sensor; the higher the density, the more reliable the estimation is generally. C: Mean Curvature, in millimeters (reciprocal), refers to the curvature of the surface passing through the located backfat measurement point P. The greater the curvature, the "sharper" the area; the smaller the curvature, the "flatter" the area. Larger sows have a more gentle back contour and a smaller curvature value C. k1: Proportional Coefficient, a constant determined through regression analysis, used to balance the contribution weights of point cloud density and curvature to backfat thickness. b1: Offset, a constant determined through regression analysis, used to correct systematic errors in the model. Specifically, feature extraction: the calculated point cloud density ρ = 0.85, and the mean curvature C = 0.15m. -1Model calculation: The preset model coefficients are k1 = 2.85 and b1 = 1.1. Results: The system determined that the backfat thickness of the sow was estimated to be 17.25 mm.

[0040] Furthermore, based on the direct mapping of backfat thickness to body condition scores, the estimated backfat thickness is mapped to a preset body condition scoring standard. For example, the correspondence between a 5-point Body Condition Score (BCS) applicable to a certain pig breed (such as Landrace) and backfat thickness is as follows: BCS 1 (Extremely Lean): The corresponding backfat thickness is less than 14.0 mm. BCS 2 (Slightly Lean): The corresponding backfat thickness is greater than or equal to 14.0 mm and less than 16.0 mm. BCS 3 (Ideal): The corresponding backfat thickness is greater than or equal to 16.0 mm and less than 20.0 mm. BCS 4 (Slightly Fat): The corresponding backfat thickness is greater than or equal to 20.0 mm and less than 22.0 mm. BCS 5 (Overfat): The corresponding backfat thickness is greater than or equal to 22.0 mm. The system obtained a backfat thickness (BFT) of 17.25 mm for a sow, with a corresponding body condition score (BCS) of 3.

[0041] An infrared thermal imaging array synchronously acquires the body surface temperature distribution, achieving a temperature resolution of 0.1 degrees Celsius and capable of identifying abnormal body temperature fluctuations exceeding 0.5 degrees Celsius. The millimeter-wave radar ranging module operates in 80GHz continuous wave mode, calculating the backfat thickness at the 10th rib of the sow by measuring the phase difference between the transmitted and echo signals, with a measurement accuracy of ±0.5mm.

[0042] In a preferred embodiment, a structured light 3D scanner projects an coded grating pattern onto the side of the sow's torso, reconstructing a 3D point cloud model of the body surface using triangulation principles. This point cloud data is then input into a pre-trained convolutional neural network model. This model, based on a ResNet-34 backbone network, is trained on a dataset containing 100,000 labeled images of sows' side profiles. It is capable of semantic segmentation of the point cloud, automatically identifying key anatomical landmarks such as the scapula, spine, and hip, and outputting a body condition score accordingly. The score ranges from 1 to 5, where 1.0-1.5 corresponds to BCS 1 (extremely lean), 1.6-2.0 corresponds to BCS 2 (slightly lean), 2.1-3.0 corresponds to BCS 3 (ideal), 3.1-4.0 corresponds to BCS 4 (slightly obese), and 4.1-5.0 corresponds to BCS 4 (slightly obese).

[0043] Body condition score is one of the key bases for selecting feeding curves.

[0044] In a preferred embodiment, a high-precision weighing platform is also included. This platform is integrated into the bottom support structure of the positioning bar and employs a four-point strain gauge sensor array with a measuring range of 0-400 kg, a resolution of 0.1 kg, and a sampling frequency of 10 Hz. To eliminate environmental vibration interference, the platform incorporates a digital filtering algorithm to perform moving average and wavelet denoising processing on the original signal. Combined with positioning from other sensors, the weighing platform ultimately outputs stable and reliable real-time weight data.

[0045] In a preferred embodiment, the inspection robot periodically (e.g., daily) collects 3D body shape data of the sow in a non-contact manner using its onboard 3D depth camera. By analyzing changes in body volume, it indirectly and accurately reflects changes in the sow's weight. There is a strong positive correlation between a sow's weight and its body volume. By establishing an accurate 3D volume model and multiplying it by a reasonable average density coefficient, the weight can be estimated. Although absolute weight may have some error, continuous weight changes (Δweight) under the same conditions can be captured with high precision. Data Acquisition and Preprocessing: When the sow is in a standard standing posture (e.g., head down eating or standing calmly), the inspection robot triggers the depth camera to acquire complete 3D point cloud data of its torso. The point cloud is then processed for noise reduction, filtering, and background segmentation to remove irrelevant point clouds such as those of the ground and fences. 3D Reconstruction and Volume Calculation: Using the processed point cloud data, a closed 3D mesh model of the sow's torso is reconstructed using Poisson reconstruction or a convex hull algorithm. Subsequently, the volume V enclosed by this 3D model is calculated. Body weight estimation model establishment: The estimated body weight (West) of a sow is related to its volume (V) and mean density (d), and can be expressed by the following basic formula: W est =V·d, where the average density d is an experimentally calibrated constant that combines the average density of the pig's muscle, fat, bone, and internal organs. Calculation of weight change: The system records the estimated weight West(t) and West(t-1) of the same sow at different dates t and t-1. The weight change ΔW is:

[0046] ΔW=W est (t)-W est (t-1)

[0047] Since systematic errors are stable in the short term and can be canceled out after subtraction, ΔW can accurately reflect the true weight change trend of sows.

[0048] In a preferred embodiment, to improve estimation accuracy, the present invention employs a more refined linear regression model that, in addition to volume, introduces features that better characterize body structure.

[0049]

[0050] Where West represents the estimated weight of the sow in kilograms. V represents the trunk volume in cubic meters, calculated using a 3D reconstruction algorithm. L BL Body length, measured in meters. It refers to the distance from the posterior border of the scapula to the ischial tuberosity, calculated in the 3D model using the Euclidean distance between corresponding key points. H SH Shoulder height, in meters. This refers to the vertical distance from the highest point of the withers to the ground, extracted directly from the 3D model. k2 and k3 are regression coefficients, constants obtained through linear regression analysis on a large sample (sows with known true weight). b2: The bias term of the regression model, also determined through regression analysis. The model considers not only the total volume V but also introduces a feature term L related to body structure. BL ·H SH 2 This is similar to an index reflecting the cross-sectional area of ​​the trunk, which allows the model to better adapt to pigs with different body structures, thereby improving the accuracy of weight estimation, especially the sensitivity to changes in body weight.

[0051] Data Acquisition: At time t, the inspection robot scans a sow to be tested and reconstructs its 3D model. The system calculates from the model: trunk volume V = 0.235 m³. 3 Body length L BL =1.45m, shoulder height H SH =0.82m, the system's preset regression coefficients are k2=450, k3=105, b2=-15. W est (t)=450×0.235+105×(1.45×0.82 2 )-15=105.75+105×(1.45×0.6724)-15=105.75+105×0.975-15=105.75+

[0052] 102.375 - 15 = 193.125 kg. The system records the sow's estimated weight on date t as 193.1 kg. The system retrieves the sow's estimated weight record for date t-1 as W. est (t-1) = 190.5 kg, change calculated: ΔW = 193.1 - 190.5 = 2.6 kg. The system determines that the sow has gained 2.6 kg in the past 24 hours. If this rate of weight gain is higher than the expected value under the current feeding curve, the system may moderately control the increase in feed intake when fine-tuning the feeding parameters to prevent the sow from becoming overweight.

[0053] All collected data—including identification code, backfat thickness, body condition score, real-time weight, ear temperature, timestamp, and field number—is uploaded by the inspection robot to the back-end management system via an industrial-grade wireless communication module.

[0054] In the curve management stage, the back-end management system's non-relational database pre-stores a standard feeding curve library, the data organization structure of which is shown in the attached figure. Figure 2 As shown, the data is stored in key-value pair format. The primary key consists of a combination of parity category and body condition score. Parity category is divided into P1 (primiparous), P2 (second parity), and Ps (third parity and above). Body condition score is divided into five intervals from BCS01 to BCS05. Each feeding curve is defined as a discrete function f(d), with the domain d being gestation age (1 ≤ d ≤ 114) or lactation age (1 ≤ d ≤ 28). The value range includes four key parameters: target feed intake T(d), warning feed intake W(d), very low feed intake L(d), and maximum feed intake M(d), all in kg / day. These parameters are calculated based on a sow nutritional requirement model and calibrated using empirical data from a large-scale pig farm over three consecutive years, covering typical requirements under different breeds, parities, and environmental conditions. See [link to relevant documentation] Figure 3 Each feeding curve sets three parameters based on different gestation ages: target feed intake, warning feed intake, and very low feed intake. The actual feed intake status of the sow is distinguished by three color blocks: green, yellow, and red. That is, the range below the very low feed intake is red, the range between very low and warning is yellow, and the range between warning and target is green.

[0055] After receiving individual data uploaded by the inspection robot, the system first parses the parity number and body condition score, matches them with the corresponding feeding curve primary key, and retrieves all parameter values ​​of that curve at the current age d. Then, the system reads the sow's weight change sequence over the past 7 days, Δw=[w_{d-6},w_{d-5},...,w_d], and calls the LinearRegression class in the Scikit-learn library to calculate the linear regression slope.

[0056]

[0057] If k < -0.3 kg / day, it is determined that the body weight loss is too rapid and there is a risk of negative energy balance. The system will increase the target feed intake T(d) by 10%. If k > +0.2 kg / day and the body condition score ≥ BCS04 (i.e., obese or overweight), it is determined that the energy intake is excessive. The system will decrease T(d) by 8%. The fine-tuned target feed intake is recorded as T'(d) and used as the core baseline parameter for feeding on that day.

[0058] In the feed control stage, the system automatically selects between quantitative and interactive feed modes based on the sow's physiological stage. Quantitative feed mode is enabled by default during the non-pregnant and gestation stages. The central controller distributes T'(d) into three meals at a ratio of 30% for breakfast, 30% for lunch, and 40% for dinner. Feeding instructions for each meal are sent to the motor driver in the corresponding pen. The driver controls the geared motor to rotate a specified number of revolutions via an encoder, driving the screw feeder to output feed of precise quality. The screw feeder has a pitch of 40mm and a diameter of 75mm. Approximately 0.1g of feed is output per motor revolution. After calibration, the system achieves feed error control within 1000g ± 15g by controlling the motor rotation speed.

[0059] During the farrowing and lactation stage, the system automatically identifies farrowing events through multimodal data fusion. When the inspection robot detects that the sow has not left the pen for 24 consecutive hours, and infrared thermal imaging shows that the temperature distribution in the abdominal area exhibits typical farrowing characteristics (such as the expansion of local hypothermia zones and signs of vasodilation), while the expected delivery date recorded by the electronic ear tag is within ±3 days, farrowing is determined to have occurred. The system automatically resets the age counter to d=1 and switches to the lactation feeding curve.

[0060] Feeding during lactation adopts a portioned, interactive feeding method. Each day is divided into 2-6 feeding periods based on the infant's age (d). For days 1-5, feeding periods are 07:00 and 14:00; for days 6-8, feeding periods are 07:00, 10:00, 14:00, and 17:00; and for days 9-28, feeding periods are 04:00, 07:00, 10:00, 14:00, 17:00, and 20:00. The maximum single feed amount C(d) for each period is set at 0.3-1.0 kg. This prevents leftover feed from accumulating and becoming moldy, while also accommodating the different feeding rates at different lactation stages. The system dynamically adjusts the feed-to-water ratio based on the sow's age (d): d = 1-3 days, feed-to-water ratio 1:1.0; d = 4-7 days, 1:1.5; d = 8-14 days, 1:2.0; d = 15-28 days, 1:2.5. The water flow is precisely controlled by a normally closed solenoid valve and flow meter, ensuring that feed and water fall into the trough simultaneously and mix naturally with the sow's eating movements.

[0061] The system supports automatic and interactive feeding modes. In automatic feeding mode, the feed trough is filled multiple times per meal according to the single feeding amount C(d), with the maximum daily feeding amount reaching the target feed intake T(d). In interactive feeding mode, only 5-10% of T'(d) is initially fed per meal, with the remainder released autonomously by the sow, maximizing the daily feed intake M(d). A mechanical triggering device is installed in front of the positioning pen, consisting of a stainless steel sleeve, a movable magnetic ring, and a Hall sensor. When the sow pushes the magnetic ring upwards by its snout, causing it to slide more than 3cm, the Hall sensor outputs a high-level pulse signal. The system records the number of triggers n per unit time (10 minutes) and reads the status m of the residual feed detection sensor (based on the resistance / capacitance / spectrum scanning principle) installed above the feed trough in real time to monitor whether there is residual feed in the trough.

[0062] The system employs a dual verification mechanism: condition 1, n ≥ N_th; condition 2, m is not triggered. The additional feed C(d) is only added when both conditions are met simultaneously. During pregnancy, N_th is dynamically adjusted based on gestational age: N_th = 2 when d < 30, N_th = 3 when 30 ≤ d < 80, and N_th = 4 when d ≥ 80. During lactation, N_th = 2, and the total daily addition does not exceed M(d). This logic effectively distinguishes between genuine feeding intentions and accidental touch behavior, avoiding ineffective feeding.

[0063] After each feeding, the system automatically initiates a water replenishment program, injecting 0.5L of clean drinking water into the feed trough. This action is controlled by a normally closed solenoid valve. The water replenishment not only meets the sow's immediate drinking needs but also uses the water flow to flush away residual feed from the inner wall of the feed trough, reducing the risk of mold and bacterial growth.

[0064] During the data analysis phase, all data, including feed intake, water consumption, weight, body condition score, backfat thickness, and trigger count, are uploaded in real time. The platform provides three standardized data views: a list view that displays each sow's ID, current age, parity, body condition score, daily target feed intake T'(d), actual feed intake A(d), feed intake achievement rate R(d) = A(d) / T'(d), water consumption, and abnormality markers in tabular form; a group view that aggregates statistics by parity or batch, calculating the group's average feed intake, a histogram of achievement rate distribution, and a weight gain trend line; and a pen card view that simulates the pigsty layout with a two-dimensional plan view, where each pen corresponds to a colored card, with the card color mapped according to the R(d) value: green (R≥0.95), yellow (0.85≤R<0.95), and red (R<0.85).

[0065] The system has a built-in screening engine that supports queries based on age range, feed intake target threshold, and abnormal water consumption (>20L or <5L per day), quickly identifying individuals requiring manual intervention. The data analysis module executes a feeding curve optimization algorithm weekly. This algorithm is based on a Bayesian optimization framework and uses the weaning weight W of piglets at 21 days of age as a benchmark. 21 To optimize the target, the target feed intake T(d) for each age group is adjusted, and the mathematical expression is as follows:

[0066]

[0067] Var(T(d))≤σ 2

[0068] Where D is the length of the lactation period (usually 28 days), C1 and C2 are the upper and lower limits of feed cost constraints (e.g., C1 = 50 kg, C2 = 70 kg), σ 2 The tolerance for feed intake fluctuations is set to 0.04. The optimization process iteratively searches for the optimal T(d) sequence using a Gaussian process surrogate model. Each iteration evaluates the correlation between weaning weight and feed intake of sows of the same parity in history. After the optimization results are generated, they must be reviewed and confirmed by veterinary experts before being updated to the standard feeding curve library.

[0069] In sow transfer scenarios, the system has automated processing capabilities. When a sow is transferred from the gestation barn to the farrowing barn, the inspection robot automatically checks the "farting status" field in her file after initially recognizing her electronic ear tag in the new pen. If this field is empty, the farrowing prediction subroutine is activated, combining the backfat thickness change rate (slope over the past 7 days) and behavioral activity index (calculated based on the movement frequency and standing time recorded by the inspection robot) to comprehensively determine whether farrowing is imminent. Once farrowing is confirmed, the system automatically creates a new lactation file, inheriting the original parity and breed information, resetting the age counter, and binding the lactation feeding curve, all without human intervention.

[0070] Example: 120 multiparous sows (breed: Large White × Landrace, parity: P2-Ps) from a large-scale pig farm were randomly divided into two groups of 60 sows each. The experimental group used the feeding method and device described in this invention, while the control group used a traditional static threshold feeding system (fixed daily feed intake of 2.8 kg during gestation, no individual variation adjustment, and no interactive feeding function). The experimental period covered the entire gestation period (114 days) and lactation period (28 days). The average daily feed intake, backfat loss, weaning-estrus interval, and 21-day weaning weight of piglets were recorded for both groups.

[0071] Comparative example: Traditional static feeding systems only set a fixed feed intake based on age, without collecting individual physiological parameters, without a dynamic adjustment mechanism, and the feeding method is timed and quantitative, without interactive triggering function.

[0072] The experimental results are shown in the table below:

[0073] Table 1 Comparison of data from the examples and comparative examples.

[0074]

[0075] Data shows that the experimental group experienced significantly increased feed intake during lactation, reduced backfat loss, shorter weaning-estrus interval, and higher weaning weight in piglets, while also substantially reducing feed waste. This fully demonstrates that the present invention effectively improves sow reproductive performance and feed utilization efficiency through dynamic sensing, precise control, and closed-loop optimization.

[0076] In summary, this invention organically integrates individual identification, multimodal physiological perception, dynamic feeding strategy generation, intelligent actuators, and data-driven optimization through engineering methods, forming a complete, reliable, and scalable precision feeding solution for sows. All technical features are clearly defined with deterministic parameters and logical rules, allowing those skilled in the art to implement this invention without obstacles and achieve the expected technical effects based on this specific embodiment.

[0077] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0078] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A feeding method for precisely regulating the feed intake of sows, characterized in that, The method comprises the following steps: Collecting individual data of the sow, the individual data comprising at least breed, parity, body condition score, backfat thickness, gestational age or farrowing age; Based on the individual data, dynamically calling a corresponding feeding curve from a pre-stored standard feeding curve library; Fine-tuning the feeding curve according to the real-time body weight change data of the sow to generate individualized feeding parameters; Based on the individualized feeding parameters, controlling the feeding equipment to perform quantitative or interactive feeding operations, and collecting real-time feed intake and water intake data; Performing big data analysis on the collected data to optimize the feeding strategy and output the feeding status prompt.

2. The feeding method as described in claim 1, characterized in that, The method comprises the following steps: identifying the individual identity of the sow through electronic ear tags; and automatically collecting at least one of backfat thickness, body condition score, body weight change, pregnancy status or farrowing status through a patrol robot.

3. The feeding method of claim 1, wherein, The standard feeding curve library stores multiple curves classified by parity and body condition score, and each curve sets at least one parameter of target feed intake, warning feed intake or very low feed intake according to gestational age or farrowing age.

4. The feeding method of claim 1, wherein, The control of the feeding equipment to perform quantitative or interactive feeding operations comprises: in the empty pregnancy stage, adopting a quantitative feeding mode to feed in stages according to meal proportion; or Adopting an interactive feeding mode to dynamically control the feeding according to the number of times the sow triggers the feeding device and the remaining amount of feed.

5. The feeding method of claim 1, wherein, The control of the feeding equipment to perform quantitative or interactive feeding operations further comprises: In the farrowing and lactation stage, setting the maximum feed intake, target feed intake, warning feed intake and very low feed intake according to the farrowing age; Adopting a meal-by-meal interactive feeding method to set the maximum feeding amount at a time; Supporting switching between wet feed and dry feed, and setting the feed-to-water ratio according to the age.

6. The feeding method of claim 1, wherein, The big data analysis on the collected data comprises: Uploading the feed intake and water intake data and the body weight change data to a cloud platform; Displaying the feeding status of each sow through a data view, the data view comprising at least one of a list view, a group view or a field card view; Sorting or filtering according to at least one parameter of feed intake compliance rate, age and water intake to locate abnormal sows.

7. The method of feeding of claim 6, wherein, The data view can batch modify the feeding parameters, including modifying at least one of age, feeding curve, plan and fasting state.

8. A sow intelligent feeding device for implementing the method according to any one of claims 1-7, characterized in that, The method comprises: A data collection module for collecting individual data of the sow; A curve management module for storing a standard feeding curve library and dynamically calling and fine-tuning the feeding curve based on the individual data; A feeding control module for controlling the feeding equipment to perform feeding operations according to individualized feeding parameters; A data analysis module for analyzing the feed intake and water intake data and outputting optimization strategies and feeding status.

9. The sow intelligent feeding device of claim 8, wherein, The data collection module comprises an electronic ear identification unit and a patrol robot data collection unit; The curve management module supports curve calling and fine-tuning according to at least one parameter of parity, body condition score and age; The feeding control module supports at least one mode of quantitative feeding, interactive feeding and wet feed control.

10. The sow intelligent feeding device of claim 8, wherein, Further comprising a human-computer interaction module for displaying feeding data through a list, group or field card view, and supporting batch operations and abnormal early warning.

Images