Intelligent precision feeding system and method for live pigs based on internet of things management and control and feeding mutual feeding

Abstract

The present application relates to the technical field of intelligent management, in particular to a pig intelligent precise feeding system and method based on internet of things management and control and feeding reciprocity. It comprises: a multi-modal data acquisition unit for collecting environmental parameters, individual characteristics, behavior videos and feeding amount; a feeding reciprocity analysis unit for real-time processing and analysis of feeding behavior videos, combining the data obtained from the feedback of the weighing type trough to obtain the real-time data of pigs, and using the real-time data of pigs to establish a feeding efficiency evaluation model of individual pigs; a feeding parameter generation unit for obtaining the optimal feeding cycle and feeding amount according to the feeding efficiency index of each pig. The system and method can more accurately evaluate the actual eating state of each pig, not only improving the accuracy of eating behavior analysis, avoiding misjudgment caused by ignoring individual differences and environmental factors, but also effectively dealing with the problem of high-frequency noise interference caused by head posture changes, improving the accuracy of eating behavior analysis.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent management, in particular to a pig intelligent precise feeding system and method based on Internet of Things management and feeding interaction. BACKGROUND

[0002] In modern animal husbandry, with the continuous expansion of breeding scale and the progress of technology, there is an increasing demand for precise monitoring and management of pig growth performance and health status. In order to meet this demand, intelligent pig raising technology has emerged, which uses Internet of Things (IoT), big data analysis, artificial intelligence (AI) and advanced sensor technology to improve breeding efficiency and animal welfare. Through the application of these technologies, real-time monitoring of pig house environment can be realized, including automatic adjustment of key parameters such as temperature, humidity and ammonia concentration, to ensure that pigs are in the most suitable living environment. At the same time, the intelligent system can also track the growth data of each pig, such as weight change, feed consumption and activity pattern, to provide scientific decision support for breeders. However, despite the significant progress made in intelligent pig raising technology, some existing solutions still have shortcomings. Existing technologies often use a unified standard to evaluate the feeding status of all pigs, ignoring the fact that pigs of different weights and sizes have different neck movement ranges and head posture changes, and thus have different feeding efficiency indexes, leading to feed waste and uneven growth. Therefore, a pig intelligent precise feeding system and method based on Internet of Things management and feeding interaction is provided. SUMMARY

[0003] The present application aims to provide a pig intelligent precise feeding system and method based on Internet of Things management and feeding interaction to solve the problem of using a unified standard to evaluate the feeding status of all pigs, ignoring the fact that pigs of different weights and sizes have different neck movement ranges and head posture changes, and thus have different feeding efficiency indexes, leading to feed waste and uneven growth.

[0004] To achieve the above-mentioned purpose, on the one hand, the present application aims to provide a pig intelligent precise feeding system based on Internet of Things management and feeding interaction, comprising:

[0005] A multi-modal data acquisition unit is used to monitor the environmental parameters of the pig house through environmental sensors, track the individual characteristics of individual pigs using RFID ear tags and body temperature sensors, capture the feeding behavior video of pigs using cameras, collect the feeding amount of pigs using weighing feeders, and time-align the environmental parameters, individual characteristics, behavior video and feeding amount;

[0006] The feeding mutual feeding analysis unit is used for real-time processing and analyzing the feeding behavior video, obtaining the real-time data of the live pig in combination with the data feedback of the weighing-type feeding trough, establishing the feeding efficiency evaluation model of each live pig by using the real-time data of the live pig, and outputting the feeding efficiency index of each live pig; the influence of the live pig weight and the feeding trough height is considered in the process of obtaining the real-time data of the live pig, and the pitch angle dynamic compensation is introduced for optimization;

[0007] The feeding parameter generation unit is used for obtaining the optimal feeding cycle and feeding amount according to the feeding efficiency index of each live pig.

[0008] As a further improvement of the technical solution, the real-time data of the live pig includes the head posture, the chewing frequency and the feeding amount, the head posture and the chewing frequency are obtained by optical flow method on the feeding behavior video, and the feeding amount is obtained by the weighing-type feeding trough.

[0009] As a further improvement of the technical solution, the specific steps for obtaining the head posture of the live pig are as follows:

[0010] S21, locating the key points of the live pig head from the video, the key points including the nose tip, the left ear, the right ear, the left side of the lower jaw and the right side of the lower jaw;

[0011] S22, taking the midpoint of the line connecting the two ears as the origin, taking the line from the nose tip to the origin as the axis, and taking the line from the left ear to the right ear as the axis; the direction perpendicular to the plane is taken as the axis;

[0012] S23, obtaining the pitch angle of each live pig at each moment:

[0013] ;

[0014] In the formula, is the pitch angle of the live pig at the time; is the origin coordinate of the live pig at the time; is the nose tip coordinate of the live pig at the time; S24, judging whether the live pig is effective feeding or not: setting the range of the effective feeding pitch angle, when the pitch angle is within the range, marking as effective feeding, otherwise marking as ineffective feeding.

[0015] S24, judging whether the live pig is effective feeding or not: setting the range of the effective feeding pitch angle, when the pitch angle is within the range, marking as effective feeding, otherwise marking as ineffective feeding.

[0016] ​​​​​​​​​​​​​As a further improvement to this technical solution, in the process of obtaining the pig's pitch angle in S23, the range of neck movement varies for pigs of different sizes, and the reference height when the pig lowers its head to eat changes with the change of feed in the trough. Therefore, the influence of pig weight and trough height is considered, specifically:

[0017] ;

[0018] In the formula, For optimized pigs exist The angle of time; For live pigs exist Weight compensation factor over time; For live pigs exist The angle of time-based feed trough height compensation;

[0019] ;

[0020] In the formula, For live pigs exist The height of feed accumulation over time; For live pigs exist The distance from the tip of time to the feeding trough;

[0021] This allows us to determine whether the pigs are effectively feeding.

[0022] Based on the pig's weight, different effective feeding tilt angle ranges are set. The body weight is within the range of pitch angles corresponding to that body weight, and continuously If all frames are within this range, it is marked as valid feeding; otherwise, it is marked as invalid feeding.

[0023] As a further improvement to this technical solution, the specific steps for obtaining the chewing frequency of pigs are as follows;

[0024] S25. Based on the left and right sides of the mandible located in S21, output the coordinates. and ,in, The coordinates are for the left side of the mandible; The coordinates of the right side of the mandible are given; and the displacement vectors of the left and right sides of the mandible are tracked using the coefficient optical flow method. Specifically:

[0025] ;

[0026] In the formula, For the image in gradient of the direction; is the image in is the gradient of the direction; is the time gradient;

[0027] S26, according to the result of S25, the displacement vector of the left and right endpoints of the mandible is obtained, and then the opening and closing amplitude is calculated based on the displacement vector of the left and right endpoints of the mandible:

[0028] ;

[0029] In the formula, is the live pig in time The opening and closing amplitude of the mandible; is the displacement vector of the left endpoint of the mandible; is the displacement vector of the right endpoint of the mandible;

[0030] S27, the opening and closing amplitude sequence of the mandible is band-pass filtered to remove high-frequency noise and low-frequency components:

[0031] ;

[0032] In the formula, is the filtered live pig in time The opening and closing amplitude sequence of the mandible; is a function of performing band-pass filtering operation; is the lowest cut-off frequency; is the highest cut-off frequency;

[0033] S28, the main frequency is extracted from the filtered opening and closing amplitude sequence of the mandible , so as to obtain the chewing frequency of the live pig:

[0034] ;

[0035] In the formula, is the live pig in time Chewing frequency; is the signal length; is the frequency index; ; is all frequency values.

[0036] As a further improvement of the technical solution, in the process of obtaining the chewing frequency, the rapid change of the pitch angle causes the generation of high-frequency noise, which further interferes with the calculation of the chewing frequency, therefore, the pitch angle dynamic compensation is introduced in the original opening and closing amplitude of the mandible , and the specific steps are as follows:

[0037] The optimized live pig is introduced into the original mandibular opening amplitude In The pitch angle of time is corrected to the mandibular opening amplitude:

[0038] ;

[0039] In the formula, The corrected mandibular opening amplitude;

[0040] According to the optimized live pig In The pitch angle of time adjusts the cutoff frequency of the band-pass filter:

[0041] ;

[0042] In the formula, The cutoff frequency of the band-pass filter; The maximum frequency adjustment amount;

[0043] Based on the adjusted cutoff frequency, the mandibular opening amplitude sequence is band-pass filtered:

[0044] ;

[0045] In the formula, The filtered optimized live pig The mandibular opening amplitude sequence at time ;

[0046] The main frequency is extracted from the filtered mandibular opening amplitude sequence to obtain the final optimized chewing frequency:

[0047] ;

[0048] In the formula, The optimized live pig The chewing frequency at time .

[0049] As a further improvement of the technical solution, the feeding mutual feedback analysis unit uses the real-time data of the live pig to establish an individual live pig feeding efficiency evaluation model, and the feeding efficiency evaluation model outputs the feeding efficiency index of each live pig. The specific steps are as follows:

[0050] Calculate the pitch angle effectiveness score, chewing frequency score and feed intake score, and take the pitch angle effectiveness score, chewing frequency score and feed intake score as the feature vector input of the feeding efficiency evaluation model:

[0051] ;

[0052] wherein, is a head pitch angle score of the live pig; is an optimal pitch angle; is an effective range of the pitch angle;

[0053] wherein, is a chewing frequency score of the live pig; is a minimum effective chewing frequency; is an optimal chewing frequency; is a maximum effective chewing frequency;

[0054]

[0055] wherein, is a feeding rate score of the live pig; is a feeding rate of the live pig; is a minimum effective feeding rate; is an optimal feeding rate; is a maximum effective feeding rate;

[0056]

[0057] wherein, is an input feature vector;

[0058] The input feature vector at the current time is processed using ReLU as the activation function to obtain a real-time feature vector, which captures the key features of the current state of the live pig:

[0059]

[0060] wherein, is a real-time feature vector; is an activation function; is a third weight matrix; is a second weight matrix; is a first weight matrix; is a first bias vector; is a second bias vector; is a third bias vector;

[0061] The data of the past time points are processed using LSTM to extract a historical feature vector reflecting the feeding behavior pattern of the live pig:

[0062]

[0063] wherein,​​​​​​​​​ For historical feature vectors; For LSTM networks;

[0064] The real-time feature vector is concatenated with the historical feature vector, and then a feeding efficiency evaluation model is constructed through a linear layer of the Sigmoid activation function, thereby generating a preliminary feeding efficiency index.

[0065] ;

[0066] In the formula, For live pigs The forage intake efficiency index; For the Sigmoid function; This is the fourth weight matrix; This is the fourth bias vector.

[0067] As a further improvement to this technical solution, a temperature compensation coefficient is introduced into the feed intake efficiency index for adjustment, generating the final feed intake efficiency index:

[0068] ;

[0069] In the formula, For live pigs Final feeding efficiency index; For live pigs In time The temperature compensation coefficient; where, , For live pigs The ambient temperature where it is located, This is the optimal temperature.

[0070] As a further improvement to this technical solution, the feeding parameter generation unit obtains the optimal feeding cycle and feeding amount as follows:

[0071] Optimal feeding cycle:

[0072] ;

[0073] In the formula, For live pigs Feeding cycle; The baseline feeding cycle; An adjustment coefficient for controlling the variation range of the feeding cycle; The target foraging efficiency index;

[0074] Optimal feeding amount:

[0075] ;

[0076] In the formula, The feeding amount of the live pig is The reference feeding amount is The reference feeding amount is The adjustment coefficient for controlling the change range of the feeding amount.

[0077] In another aspect, the present application provides a live pig intelligent precision feeding method based on Internet of Things management and foraging mutual feeding, which is used in the live pig intelligent precision feeding system based on Internet of Things management and foraging mutual feeding, and includes the following steps:

[0078] S1, monitoring the environmental parameters of the pig house through the environmental sensor, tracking the individual characteristics of the individual live pig by using the RFID ear tag and the body temperature sensor, capturing the foraging behavior video of the live pig by means of the camera, and collecting the foraging amount of the live pig by using the weighing type trough;

[0079] S2, real-time processing and analyzing the foraging behavior video, combining the data fed back by the weighing type trough to obtain the real-time data of the live pig, and using the real-time data of the live pig to establish the foraging efficiency evaluation model of each live pig, and the foraging efficiency evaluation model outputs the foraging efficiency index of each live pig;

[0080] S3, obtaining the optimal feeding cycle and the feeding amount according to the foraging efficiency index of each live pig.

[0081] Compared with the prior art, the present application has the following beneficial effects:

[0082] 1. In the live pig intelligent precision feeding system and method based on Internet of Things management and foraging mutual feeding, the differences in the neck movement range of live pigs with different weights and the influence of the height change of the feed in the trough on the reference height when feeding with the head lowered are considered, so as to ensure the effective recognition of the head posture, the system can more accurately evaluate the actual feeding state of each pig, not only improve the accuracy of the feeding behavior analysis, avoid the misjudgment caused by ignoring individual differences and environmental factors, but also further optimize the feeding strategy, and improve the breeding efficiency and animal welfare.

[0083] 2. In the live pig intelligent precision feeding system and method based on Internet of Things management and foraging mutual feeding, in order to accurately obtain the chewing frequency, a pitch angle dynamic compensation mechanism is introduced to correct the original mandibular opening amplitude, which can effectively deal with the high-frequency noise interference problem caused by the change of the head posture, improve the accuracy of the feeding behavior analysis, and ensure that the real chewing activity can be reliably recognized even in complex feeding environment. BRIEF DESCRIPTION OF DRAWINGS

[0084] Figure 1 The figure is the overall flow chart of the present application;

[0085] Figure 2 The figure is the overall method flow chart of the present application;

[0086] The meanings of the various numbers in the figure are:

[0087] 1, multi-modal data acquisition unit; 2, foraging interaction analysis unit; 3, feeding parameter generation unit. DETAILED DESCRIPTION

[0088] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor are within the scope of protection of the present application.

[0089] Embodiment 1: Please refer to Figure 1 As shown, an intelligent precision feeding system for live pigs based on Internet of Things management and control and foraging interaction is provided, including a multi-modal data acquisition unit 1, a foraging interaction analysis unit 2 and a feeding parameter generation unit 3.

[0090] The multi-modal data acquisition unit 1 is used to monitor the environmental parameters of the pig house through the environmental sensor, track the individual characteristics of individual live pigs using RFID ear tags and body temperature sensors, capture the foraging behavior video of live pigs with the help of the camera, collect the foraging amount of live pigs using the weighing type trough, and time align the environmental parameters, individual characteristics, behavior video and foraging amount.

[0091] Specifically, the environmental sensor includes a digital temperature and humidity sensor (such as SHT35), an electrochemical ammonia sensor (such as MQ-137), and an illumination intensity sensor. 3-5 nodes are arranged at the top of each pigpen to form a grid monitoring network, with a sampling frequency of 1 time / minute and an accuracy of ±0.5℃ (temperature), ±2%RH (humidity), ±5ppm (ammonia). The RFID ear tag is a ultra-high frequency RFID ear tag (ISO18000-6C protocol), and the body temperature sensor is a patch type body temperature sensor (accuracy ±0.1℃). The ear tag and the body temperature sensor are integrated through a flexible circuit and transmitted to the base station through LoRa wireless transmission. The camera is a wide-angle infrared camera (resolution 1920x1080@30fps) with a built-in waterproof and anti-fog optical cover. YOLOv5s is used for real-time detection of live pig heads, and the ROI region is cropped and input into the analysis model. The structure of the weighing type trough is a cantilever beam type weighing sensor (range 0-50kg, accuracy ±2g), with a stainless steel moisture-proof structure and a timestamp aligned with the camera, recording weight changes every 0.5 seconds. A denoising module is provided in the weighing type trough, which eliminates noise data caused by live pig stepping vibration based on Kalman filtering algorithm.

[0092] Feeding feedback analysis unit 2 is used to process and analyze feeding behavior videos in real time. It combines the data from the weighing trough feedback to obtain real-time data of pigs. It uses the real-time data of pigs to establish a feeding efficiency evaluation model for each pig. The feeding efficiency evaluation model outputs the feeding efficiency index of each pig. In the process of obtaining real-time data of pigs, the influence of pig weight and trough height is considered and dynamic compensation of pitch angle is introduced for optimization.

[0093] Real-time data of pigs includes head posture, chewing frequency, and feed intake. Head posture and chewing frequency are obtained by optical flow method from videos of feeding behavior, and feed intake is obtained by weighing feed troughs.

[0094] Feed intake was obtained using a weighing feed trough, specifically by subtracting the weight of the feed trough after the pigs had eaten from the weight of the feed trough before eating. The instantaneous feed intake is calculated, and then the instantaneous feed intake of all data points within the statistical period is summed to obtain the total feed intake for a given period.

[0095] The specific steps for obtaining the head posture of a pig are as follows;

[0096] S21. Locate the key points of the pig's head from the video. The key points include the tip of the nose, the left ear, the right ear, the left side of the lower jaw, and the right side of the lower jaw.

[0097] S22. Mark the midpoint of the line connecting the two ears as the origin, and the distance from the tip of the nose to the origin... The connection as The axis, the line connecting the left ear to the right ear is denoted as . Axis; perpendicular to The direction of the plane is denoted as axis;

[0098] S23, Obtain each pig Pitch angle at any moment:

[0099] ;

[0100] In the formula, For live pigs exist The angle of time; For live pigs exist The origin coordinates of time; For live pigs exist The coordinates of time at the tip of the nose; For live pigs exist Left ear coordinate of time; For live pigs exist The right ear coordinate of time;

[0101] Pitch angle represents the up-and-down tilt angle of the pig's head, reflecting its eating posture. When the head is tilted downward, it indicates that the pig is approaching the feeding trough and may be eating. When the head is raised, it indicates that the pig may be observing or resting.

[0102] S24, determine whether the pig is effectively eating: set the range of effective eating pitch angle, when within the range of pitch angle, it is marked as effective eating, otherwise it is marked as ineffective eating; ineffective eating such as observation, head turning or rest, etc. to exclude the case that the pig is near the feeding trough but not eating, such as sniffing, playing or social behavior, only record data when the pig is actually eating, improve the accuracy of feed intake calculation, avoid false triggering of the feeding system due to non-eating state of the pig.

[0103] In the process of obtaining the pitch angle of the pig in S23, the neck movement range of pigs of different body types is different, and the reference height of the pig when lowering its head to eat changes with the change of the feed in the feeding trough. Specifically, pigs with larger body weight usually have shorter necks, smaller head movement ranges, and smaller pitch angle change amplitudes; pigs with smaller body weight have more flexible necks, larger head movement ranges, and larger pitch angle change amplitudes. If the change of the feed height is not considered, the eating state of the pig may be misjudged, therefore, the influence of the pig's body weight and the feeding trough height is considered, specifically:

[0104] ;

[0105] In the formula, is the optimized pig pitch angle at time; is the pig body weight compensation coefficient at time; pigs with larger body weight have smaller head movement ranges, and the pitch angle change amplitude is reduced by ; pigs with smaller body weight have larger head movement ranges, and the pitch angle change amplitude is increased by ; is the pig feeding trough height compensation angle at time;

[0106] ;

[0107] In the formula, is the pig feed accumulation height at time; when the feed in the feeding trough decreases, the feed height decreases; when the feed in the feeding trough increases, the feed height increases; is the pig pitch angle compensation coefficient at The distance from the tip of the nose to the feeding trough (calibrated by the camera); when the height of the feeding trough changes (such as feed accumulation), the pitch angle reference value is automatically adjusted;

[0108] Further determine whether the pig is effective feeding:

[0109] According to the body weight of the pig, set different effective feeding pitch angle ranges, when the pig The body weight is within the pitch angle range corresponding to the body weight, and the body weight is within the range for a continuous frame, it is marked as effective feeding, otherwise it is marked as ineffective feeding; avoid misjudgment due to weight difference: the angle change of the pig with large weight is small, so set a more relaxed range. The angle change of the pig with small weight is large, so set a more strict range. Avoid misjudgment caused by temporary head swing: continuous frame judgment can filter noise and prevent misjudgment as feeding.

[0110] Considering the difference in neck movement range of pigs of different weights and the influence of feed height change in the feeding trough on the reference height when feeding, the system can more accurately evaluate the actual feeding state of each pig. Specifically, the system integrates pig weight information and real-time monitoring of feeding trough height to dynamically adjust the calculation of the pitch angle, ensuring effective recognition of head posture. This personalized monitoring method not only improves the accuracy of feeding behavior analysis, avoiding misjudgment caused by ignoring individual differences and environmental factors, but also further optimizes feeding strategies, improving breeding efficiency and animal welfare.

[0111] The specific steps of obtaining the pig's chewing frequency are as follows:

[0112] S25, according to the left and right sides of the mandible positioned in S21, output coordinates and , wherein, is the left side coordinate of the mandible; is the right side coordinate of the mandible; and the coefficient optical flow method is used to track the displacement vector of the left and right sides of the mandible, specifically:

[0113] ;

[0114] In the formula, is the gradient of the image in the direction; is the gradient of the image in the direction; is the time gradient;

[0115] S26, according to the displacement vector of the left and right endpoints of the mandible obtained in S25, further calculate the opening amplitude based on the displacement vector of the left and right endpoints of the mandible:

[0116] ;

[0117] wherein, is the opening amplitude of the mandible of the live pig at time ; is the displacement vector of the left endpoint of the mandible; is the displacement vector of the right endpoint of the mandible; if the mandible is opened, the distance between the left and right endpoints increases; if closed, the distance between the left and right endpoints decreases.

[0118] S27, the opening amplitude sequence of the mandible is band-pass filtered to remove high-frequency noise and low-frequency components:

[0119] ;

[0120] wherein, is the opening amplitude sequence of the mandible of the live pig at time after filtering, the frequency band of 0.5-3 Hz is retained, because the chewing frequency of the live pig is usually within the range of 0.5-3 Hz, and signals beyond this range may be noise or other irrelevant actions; is a function for performing band-pass filtering operation; is the lowest cut-off frequency; is the highest cut-off frequency; the parameters are signal, lowest cut-off frequency (0.5 Hz) and highest cut-off frequency (3 Hz) respectively;

[0121] S28, the main frequency is extracted from the filtered opening amplitude sequence of the mandible, so as to obtain the chewing frequency of the live pig:

[0122] ;

[0123] wherein, is the chewing frequency of the live pig at time ; is the length of the signal; is the frequency index; ; the frequency component with the largest amplitude in the Fourier transform result is selected as the main chewing frequency of the live pig; is all frequency values, indicating a specific frequency value with the largest amplitude among all frequencies;

[0124] ​​​​​During the process of chewing frequency acquisition, the rapid change of pitch angle leads to the generation of high frequency noise, which interferes with the calculation of chewing frequency. Specifically, when the head is raised (such as eating high feed), the projection of the mandibular movement is reduced, resulting in the underestimation of the chewing amplitude; when the head is horizontal, the data of the mandibular opening and closing amplitude is true and reliable; when the head is fully raised (such as licking food, observation), the mandibular opening and closing almost loses practical significance, and the affected data needs to be removed; therefore, the pitch angle dynamic compensation is introduced into the original mandibular opening and closing amplitude , and the specific steps are as follows:

[0125] The optimized pig 's pitch angle at the time of is introduced into the mandibular opening and closing amplitude:

[0126] ;

[0127] In the formula, is the corrected mandibular opening and closing amplitude;

[0128] When the pig's head is raised , the projection of the actual mandibular movement amplitude in the vertical direction is reduced, and the attenuation is performed through the cosine function; when (horizontal posture), the compensation coefficient is 1, and there is no attenuation; when (completely raised head), the compensation coefficient is 0, and the invalid data is completely removed.

[0129] Then, the pitch angle of the pig at the time of is used to adjust the cutoff frequency of the band-pass filter to suppress the high frequency noise introduced by the abnormal head posture:

[0130] ;

[0131] In the formula, is the cutoff frequency of the band-pass filter; is the maximum frequency adjustment amount; when (obviously raised), the high cutoff frequency is increased to suppress the high frequency signal caused by the abnormal posture; otherwise, it remains 3Hz, which does not over-adjust to ensure the analysis of normal eating behavior; in this way, the high frequency interference caused by the change of head posture can be effectively suppressed, and the filtered mandibular movement signal is more stable;

[0132] Based on the adjusted cutoff frequency, the mandibular opening and closing amplitude sequence is subjected to band-pass filtering:

[0133] ;

[0134] In the formula, is the pig the mandibular opening amplitude sequence at time ;

[0135] extract the dominant frequency from the filtered mandibular opening amplitude sequence ;

[0136] ;

[0137] wherein, the optimized piglet mastication frequency at time ;

[0138] This compensation mechanism can effectively deal with the problem of high-frequency noise interference caused by head posture changes, especially when the piglet's head is raised or completely tilted up, which may lead to underestimated mastication amplitude or data distortion due to the reduction or loss of actual significance of the mandibular movement projection. By monitoring and adjusting the compensation factor calculated according to the pitch angle in real time, the system can provide more accurate mastication frequency estimation during different head positions, such as feeding high feed, normal horizontal feeding, and head-up observation.

[0139] The feed intake mutual feedback analysis unit 2 uses the real-time data of the piglets and the weight change curve to establish a feed intake efficiency evaluation model for individual piglets, and the feed intake efficiency evaluation model outputs the feed intake efficiency index of each piglet. The specific steps are as follows:

[0140] Calculate the pitch angle effectiveness score, mastication frequency score, and feed intake score, and use the pitch angle effectiveness score, mastication frequency score, and feed intake score as the feature vector input of the feed intake efficiency evaluation model:

[0141] ;

[0142] wherein, is the head pitch angle score of the piglet ; is the optimal pitch angle; is the effective range of the pitch angle; this index is used to measure whether the head posture of the piglet during feeding is appropriate. By comparing the actual observed pitch angle with the ideal angle and considering an allowed angle range, a score between 0 and 1 can be obtained, indicating the effectiveness of the current feeding posture of the piglet, which helps to identify those piglets that may not be able to feed normally due to physical discomfort or environmental problems.

[0143] ;

[0144] wherein, is the mastication frequency score of the piglet ; is the minimum effective chewing frequency; is the optimal chewing frequency; is the maximum effective chewing frequency; the actual chewing frequency of the live pig is compared with the optimal value, and the score is determined by setting the minimum effective frequency and the maximum effective frequency. Such a scoring mechanism can help monitor the health status of the live pig, because abnormal chewing frequency can be an early sign of disease or other health problems;

[0145] ;

[0146] wherein, is the live pig is the feed intake score per unit time; is the live pig is the feed intake per unit time; is the minimum effective feed intake; is the optimal feed intake; is the maximum effective feed intake; the score reflects the appetite of the live pig in a specific period of time. By comparing the actual feed intake rate with the ideal minimum, maximum and optimal feed intake rate, a score reflecting the health degree of the live pig's feeding behavior can be obtained;

[0147] ;

[0148] wherein, is the input feature vector;

[0149] The input feature vector at the current time is processed using ReLU as the activation function to obtain the real-time feature vector, which captures the key features of the current state of the live pig:

[0150] ;

[0151] wherein, is the real-time feature vector; is the activation function; is the third weight matrix; is the second weight matrix; is the first weight matrix; is the first bias vector; is the second bias vector; is the third bias vector; wherein, ; , , ; , , ;

[0152] The past Data at a time point, extract the historical feature vector reflecting the pig feeding behavior pattern:

[0153] ;

[0154] In the formula, is a historical feature vector; is an LSTM network;

[0155] The real-time feature vector and the historical feature vector are spliced, and then a foraging efficiency evaluation model is constructed through a linear layer of Sigmoid activation function, thereby generating a preliminary foraging efficiency index:

[0156] ;

[0157] In the formula, is the foraging efficiency index of the pig ; ; is a Sigmoid function that maps the output to the range; is a fourth weight matrix; is a fourth bias vector;

[0158] 3-layer fully connected network (64→32→16 nodes, ReLU activation), processing real-time features ; LSTM network (16 units), processing time series features of pigs (e.g., the last 1 hour of sequence), concatenating the two-channel output, and outputting the foraging efficiency index of the pig through the Sigmoid function;

[0159] In summary, the model construction part uses a three-layer fully connected neural network (FCN) containing ReLU activation function to process features obtained from three different dimensions: pitch effectiveness score, chewing frequency score, and foraging amount score. In addition, an LSTM network is used to capture dynamic changes over time, i.e., the feeding pattern over a period of time. After combining these two feature vectors, the Sigmoid function is used to map them to the [0, 1] interval to form the foraging efficiency index;

[0160] A temperature compensation coefficient is introduced to adjust the foraging efficiency index, generating the final foraging efficiency index:

[0161] ;

[0162] In the formula, is the final foraging efficiency index of the pig ; is the pig In time Temperature compensation coefficient;

[0163] in, , For live pigs The ambient temperature The optimal temperature;

[0164] The feed intake efficiency assessment model can output a feed intake efficiency index (0-1, 1 represents the best state) by taking into input head pitch angle (to determine whether feeding is effective), chewing frequency (to assess feeding speed) and feed intake per unit time (g / s). When a pig is detected to have a feed intake efficiency of <0.5 for 3 consecutive times, an alarm is triggered and a camera is linked to capture video clips for manual review of health status.

[0165] Feeding parameter generation unit 3 is used to obtain the optimal feeding cycle and feeding amount based on the feed intake efficiency index of each pig;

[0166] Optimal feeding cycle:

[0167] ;

[0168] In the formula, For live pigs The feeding cycle; The baseline feeding cycle is set at four hours. An adjustment coefficient for controlling the variation range of the feeding cycle; The target foraging efficiency index;

[0169] when This indicates that the pig is currently eating well and may not require frequent feeding. Therefore, the feeding cycle should be appropriately extended to avoid overfeeding. If the pig is in a critical condition, it indicates that the pig may have health problems or need more nutrition. In this case, the feeding cycle should be shortened to ensure that the pig receives enough feed.

[0170] Optimal feeding amount:

[0171] ;

[0172] In the formula, For live pigs The amount of feed given; The baseline feeding amount; An adjustment coefficient for controlling the variation in feeding amount;

[0173] If the feed intake efficiency index of a pig is higher than the target value, the feeding amount can be appropriately reduced, because it indicates that the pig has reached a better growth state and does not need additional feed supply. Conversely, if the feed intake efficiency index is lower than the target value, the feeding amount needs to be increased to meet the higher nutritional needs of the pig and help it recover to the ideal state.

[0174] Through the above two formulas, the intelligent feeding system can realize personalized management of each pig, not only improving feed conversion rate and reducing breeding cost, but also effectively improving the growth rate and health level of pigs. Embodiment 2

[0175] Please refer to Figure 2 The present embodiment provides a pig intelligent precise feeding method based on Internet of Things management and control and feed intake mutual feeding, which is used for the pig intelligent precise feeding system based on Internet of Things management and control and feed intake mutual feeding as described above, and includes the following steps:

[0176] S1, monitor the environmental parameters of the pig house through the environmental sensor, track the individual characteristics of the individual pig by using the RFID ear tag and the body temperature sensor, capture the feed intake behavior video of the pig by means of the camera, and collect the feed intake amount of the pig by using the weighing type trough;

[0177] S2, real-time processing and analysis of the feed intake behavior video, combined with the data obtained from the weighing type trough to obtain the real-time data of the pig, and using the real-time data of the pig to establish a feed intake efficiency evaluation model for each pig, the feed intake efficiency evaluation model outputs the feed intake efficiency index of each pig;

[0178] S3, according to the feed intake efficiency index of each pig, the optimal feeding cycle and the feeding amount are obtained.

[0179] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above embodiments, and the above embodiments and descriptions in the specification are only preferred examples of the present application and are not intended to limit the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the present application.

Claims

1. A pig intelligent precision feeding system based on Internet of Things management and control and feeding mutual feeding, characterized in that, Comprise: A multi-modal data acquisition unit (1) for monitoring environmental parameters of a pig house through environmental sensors, tracking individual characteristics of individual piglets with RFID ear tags and body temperature sensors, capturing piglet feeding behavior videos with cameras, collecting piglet feed intake using a weighable feeding trough, and time-aligning the environmental parameters, individual characteristics, behavior videos, and feed intake; A feeding interaction analysis unit (2) for real-time processing and analysis of feeding behavior videos, obtaining real-time data of piglets in combination with data feedback from the weighable feeding trough, wherein the real-time data of piglets includes head posture, chewing frequency, and feed intake, the head posture and chewing frequency are obtained by optical flow method on the feeding behavior video, and the feed intake is obtained by the weighable feeding trough; a feeding efficiency evaluation model for each piglet is established using the real-time data of piglets, the feeding efficiency evaluation model outputs the feeding efficiency index of each piglet; the influence of piglet body weight and feeding trough height is considered during the process of obtaining the real-time data of piglets, and a pitch angle dynamic compensation is introduced for optimization; Wherein, considering the influence of piglet body weight and feeding trough height, specifically: ； In the formula, Optimized live pig In Pitch angle of time; Live pig In Body weight compensation coefficient of time; Live pig In Feed trough height compensation angle of time; ； In the formula, for live pigs At Time of feed pile height; for live pigs At Time of snout to trough distance; According to the body weight of the live pig, different effective feeding pitch angle ranges are set, when the body weight of the live pig is within the pitch angle range corresponding to the body weight, and the live pig is continuously within the range in each frame, it is marked as effective feeding, otherwise it is marked as ineffective feeding; The specific steps of obtaining the piglet head posture are as follows: S21, locate the piglet head key points from the video, the key points include nose tip, left ear, right ear, left side of lower jaw, and right side of lower jaw; S22. Mark the midpoint of the line connecting the two ears as the origin, and the distance from the tip of the nose to the origin... The connection as The axis, the line connecting the left ear to the right ear is denoted as . Axis; perpendicular to The direction of the plane is denoted as axis; S23, acquiring each live pig pitch angle at the moment: ； wherein for live pigs at time of pitch angle; for live pigs at time of origin coordinates; for live pigs at time of tip coordinates; Determine whether the pig is effective feeding: set the range of effective feeding pitch angle, when In the range of the pitch angle, it is marked as effective feeding, otherwise it is marked as ineffective feeding; And the specific steps of obtaining the piglet chewing frequency are as follows: S25, outputting coordinates according to the left and right sides of the lower jaw positioned in S21 and wherein, is the left side of the lower jaw coordinate; is the right side of the lower jaw coordinate; and a displacement vector of the left and right sides of the lower jaw is tracked by using a coefficient optical flow method , specifically: ； wherein is the gradient of the image in direction; is the gradient of the image in direction; is the temporal gradient; S26, get the displacement vector of the left and right endpoints of the lower jaw according to the result of S25, and then calculate the opening amplitude based on the displacement vector of the left and right endpoints of the lower jaw: ； wherein is the live pig at time is the mandibular opening amplitude at time is the displacement vector of the left end point of the mandible is the displacement vector of the right end point of the mandible S27, sequence of mandibular opening-closing amplitudes Band-pass filtering is performed to remove high-frequency noise and low-frequency components: ； wherein is the filtered live pig is the sequence of mandibular opening amplitude at time is the sequence of mandibular opening amplitude at time is a function that performs a band-pass filtering operation; is the lowest cut-off frequency; is the highest cut-off frequency; S28, extracting the dominant frequency from the filtered mandibular opening amplitude sequence extracting the dominant frequency from the filtered mandibular opening amplitude sequence, thereby obtaining the chewing frequency of the live pig ； wherein is the live pig at time chewing frequency; is the signal length; is the frequency index; ; is all frequency values; In the original mandibular opening amplitude The specific steps are as follows: Introducing optimized live pig in original mandibular opening amplitude In Time's pitch angle gets corrected mandibular opening amplitude: ； In the formula, is the corrected mandibular opening amplitude; According to the optimized live pigs In The cut-off frequency of the pitch angle adjustment band-pass filter: ； In the formula, is the cut-off frequency of the bandpass filter; is the maximum frequency adjustment amount; Bandpass filtering the mandibular opening amplitude sequence based on the adjusted cutoff frequency : ； In the formula, live pigs after filtering optimization at time the mandibular opening amplitude sequence; from the filtered mandibular opening amplitude sequence extracting dominant frequencies therefrom, obtaining a final optimized mastication frequency: ； In the formula, Optimized live pigs At time Chewing frequency; A feeding parameter generation unit (3) for obtaining the optimal feeding period and feeding amount according to the feeding efficiency index of each piglet.

2. The intelligent precision feeding system for live pigs based on Internet of Things management and feeding mutual feeding according to claim 1, characterized in that: In the feeding interaction analysis unit (2), a feeding efficiency evaluation model for individual piglets is established using the real-time data of piglets, the feeding efficiency evaluation model outputs the feeding efficiency index of each piglet, the specific steps are as follows: computing a pitch angle effectiveness score , a chewing frequency score and a feed intake score and using the pitch angle effectiveness score, the chewing frequency score and the feed intake score as a feature vector input to a feed efficiency assessment model ; Wherein, ； Use ReLU as the activation function to process the feature vector of the current input to obtain the real-time feature vector, which captures the key features of the current state of the piglet: ； wherein, is a real-time feature vector; is an activation function; is a third weight matrix; is a second weight matrix; is a first weight matrix; is a first bias vector; is a second bias vector; is a third bias vector; LSTM is used to process data at past time points to extract a history feature vector reflecting the pig's eating behavior pattern: ； In the formula, is a historical feature vector; is an LSTM network; Concatenate the real-time feature vector with the historical feature vector, then construct the feeding efficiency evaluation model through the linear layer of the Sigmoid activation function, thereby generating the preliminary feeding efficiency index: ； wherein is a feed efficiency index of the pig; is a feed efficiency index of the pig; is a Sigmoid function; is a fourth weight matrix; is a fourth bias vector.

3. The intelligent precision feeding system for pig based on Internet of Things management and feeding inter-donation according to claim 2, characterized in that: Introduce a temperature compensation coefficient into the feeding efficiency index for adjustment to generate the final feeding efficiency index: ； wherein live pigs final feed efficiency index; live pigs at time temperature compensation factor; wherein, , live pigs ambient temperature in which the live pigs are located, optimum temperature.

4. The intelligent precision feeding system for breeding pigs based on Internet of Things management and feeding mutual feeding according to claim 3, characterized in that, The feeding parameter generation unit (3) obtains the optimal feeding period and feeding amount, specifically: Optimal feeding period: ； In the formula, For live pigs The feeding cycle; The baseline feeding cycle; An adjustment coefficient for controlling the variation range of the feeding cycle; The target foraging efficiency index; Optimal feeding amount: ； In the formula, For live pigs The amount of feed given; The baseline feeding amount; An adjustment coefficient used to control the variation in feeding amount.

5. The intelligent precision feeding method for live pigs based on the Internet of Things management and control and feeding mutual feeding, used in the intelligent precision feeding system for live pigs based on the Internet of Things management and control and feeding mutual feeding as claimed in any one of claims 1-4, characterized in that, Comprise the following steps: S1, monitor the environmental parameters of the pig house through environmental sensors, track the individual characteristics of individual piglets with RFID ear tags and body temperature sensors, capture the feeding behavior videos of piglets with cameras, and collect the feed intake of piglets using a weighable feeding trough; S2, real-time processing and analysis of feeding behavior videos, obtaining real-time data of piglets in combination with data feedback from the weighable feeding trough, establishing a feeding efficiency evaluation model for each piglet using the real-time data of piglets, the feeding efficiency evaluation model outputs the feeding efficiency index of each piglet; S3, obtaining the optimal feeding period and feeding amount according to the feed efficiency index of each piglet.

Images