Animal husbandry and veterinary intelligent diagnosis system based on machine learning

Abstract

The invention discloses an intelligent animal husbandry and veterinary diagnosis system based on machine learning, and particularly relates to the technical field of intelligent animal husbandry diagnosis. The livestock body surface temperature, the livestock chewing data and the livestock movement data are processed to obtain the real body temperature, the rumination time, the movement state and the movement speed of the livestock, and the livestock heart rate, the respiratory rate, the real body temperature, the rumination time and the movement speed are input into a livestock sign abnormity recognition model for abnormity recognition. The behavior data of the livestock is obtained through the multi-angle camera, the abnormal image data set is marked after a series of processing, the received livestock sign abnormal data set and the received livestock behavior abnormal image data set are input into the livestock health intelligent diagnosis model for health diagnosis, and the livestock illness state diagnosis efficiency and accuracy can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of livestock intelligent diagnosis, and more specifically, to a livestock veterinary intelligent diagnosis system based on machine learning. Background Art

[0002] Traditional livestock health diagnosis mainly relies on the sensory observation and experience judgment of veterinary personnel. These methods are simple and easy to implement, and can be carried out in any place. However, the limitations of traditional livestock health diagnosis are also obvious, such as insufficient diagnostic accuracy, low diagnostic efficiency, limited diagnostic scope, and inability to meet the needs of large-scale farming. Therefore, it is necessary to develop a more advanced and intelligent livestock diagnosis method to improve the diagnostic efficiency and accuracy.

[0003] The existing livestock health diagnosis monitors the exercise physiological data of livestock through a livestock physiological index intelligent monitoring system composed of multiple physiological index collection devices, wireless base stations, servers, etc., and establishes a standard physiological index model and a health index of livestock through the exercise physiological data. During the monitoring process, the health indexes of livestock that do not conform to the model rules and are abnormal are sent to the veterinary side to assist veterinarians in diagnosing the diseases of livestock. This greatly reduces the cumbersome work of diagnosing livestock one by one, accurately locks the scope of livestock with abnormalities, and can improve the diagnostic efficiency and accuracy.

[0004] Although the existing livestock health diagnosis system can improve the diagnostic efficiency and accuracy to a certain extent, there are still the following problems: First, the current optimized livestock AI judgment is mainly based on the comparative difference analysis between the exercise physiological data of livestock monitored by intelligent monitoring devices and the established model. However, when livestock are ill, there may be multiple abnormal physical sign parameters or behavior data. The monitoring scope of the existing system is relatively narrow, and it cannot provide sufficient data support for veterinarians' diagnosis, and the accuracy of disease diagnosis still needs to be improved; Second, due to the limited experience, veterinarians need to exclude irrelevant causes one by one when receiving abnormal monitoring results. Due to the insufficient monitoring indicators, the process of excluding and tracing the causes may consume a lot of time, which is not conducive to the subsequent treatment of livestock, and the efficiency of cause diagnosis still needs to be further improved. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a livestock veterinary intelligent diagnosis system based on machine learning, which collects the physical sign data and behavior data of livestock, respectively performs preprocessing and anomaly analysis, and simultaneously constructs a health intelligent diagnosis model. The abnormal physical sign data group and the abnormal behavior image data group of livestock can be directly input into the trained and tested livestock health intelligent diagnosis model as input terminals for health diagnosis, which can effectively solve the problems of low diagnostic efficiency and low diagnostic accuracy.

[0006] To achieve the above object, the present invention provides the following technical solution: A livestock veterinary intelligent diagnosis system based on machine learning, comprising: Vital sign and behavior data acquisition module: Collect the vital sign data of livestock through multiple sensors configured on the intelligent collar, and obtain the behavior data of livestock through multi-angle cameras; Vital sign data preprocessing module: Process the body surface temperature, chewing data, and movement data of livestock to obtain the real body temperature, rumination time, movement state, and movement speed of livestock; Behavior data preprocessing module: Use image processing tools to extract the feeding posture images, feeding action images, movement posture images, and lying posture images of livestock from the behavior data, and identify whether the extracted posture images are accurate. Update the behavior data based on the recognition results; Vital sign abnormality analysis module: Input the heart rate, respiratory rate, real body temperature, rumination time, and movement speed of livestock into the trained and tested livestock vital sign abnormality recognition model for abnormality recognition, mark the vital sign abnormal data group, and after confirming the vital sign abnormality of livestock, send the vital sign abnormal data group to the health intelligent diagnosis module; Behavior abnormality recognition module: Calculate the proportion of correct postures of eating behavior, movement behavior, and rest behavior, then calculate the abnormality coefficient of each behavior, and then calculate the comprehensive behavior abnormality index, mark the abnormal image data group, and after confirming the behavior abnormality of livestock, send the livestock behavior abnormal image data group to the health intelligent diagnosis module; Health intelligent diagnosis module: Input the received livestock vital sign abnormal data group and livestock behavior abnormal image data group as inputs into the trained and tested livestock health intelligent diagnosis model for health diagnosis, output the abnormal cause with the highest matching degree and the corresponding treatment plan, determine whether the model diagnosis result triggers a secondary diagnosis, if it triggers, perform a secondary diagnosis and determine the final output abnormal cause and treatment plan based on the set rules, if it does not trigger, directly output the abnormal cause and treatment plan diagnosed by the model; Database: Used to store the data information of all modules in the system.

[0007] Technical effects and advantages of the present invention: The present invention collects the vital sign data of livestock through multiple sensors configured on the intelligent collar, processes the body surface temperature, chewing data, and movement data of livestock to obtain the real body temperature, rumination time, movement state, and movement speed of livestock, inputs the heart rate, respiratory rate, real body temperature, rumination time, and movement speed of livestock into the trained and tested livestock vital sign abnormality recognition model for abnormality recognition, marks the vital sign abnormal data group, and outputs the vital sign abnormal data group after confirming the vital sign abnormality of livestock, improving the accuracy and recognition efficiency of vital sign abnormality recognition.

[0008] The present invention obtains the behavioral data of livestock through multi-angle cameras, uses image processing tools to extract the feeding posture images, feeding action images, movement posture images, and lying posture images of livestock from the behavioral data, and identifies whether the extracted posture images are accurate. Based on the recognition results, the behavioral data is updated. After calculating the correct posture ratios of feeding behavior, movement behavior, and rest behavior, the abnormality coefficient of each behavior is calculated, and then the comprehensive behavior abnormality index is calculated. The abnormal image data group is marked, and after confirming that the livestock behavior is abnormal, the livestock behavior abnormal data group is output, improving the accuracy and recognition efficiency of behavior abnormality recognition.

[0009] After receiving the livestock physical sign abnormal data group and the livestock behavior abnormal image data group, the present invention inputs the received livestock physical sign abnormal data group and the livestock behavior abnormal image data group as input terminals into the livestock health intelligent diagnosis model that has been trained and passed the test for health diagnosis. The model automatically matches the physical sign abnormal parameters and the behavior abnormal images with the abnormal physical sign manifestations and abnormal behavior manifestations corresponding to different etiologies in the machine learning process one by one. The ratio of the number of overlapping abnormal parameters to the actual number of abnormal parameters of the livestock is the matching degree. The abnormal etiology with the highest matching degree and the corresponding treatment plan are output as the output terminal. Then, the diagnosis result is determined, that is, the abnormal etiology matching degree is compared with the set matching degree confidence threshold. If the abnormal etiology matching degree is greater than or equal to the set matching degree confidence threshold, the abnormal etiology with the highest matching degree and the corresponding treatment plan can be directly sent to the livestock user terminal. Otherwise, information is sent to multiple veterinarians registered in the system to request a second diagnosis. If the second diagnosis result is consistent with the model diagnosis result, the abnormal etiology with the highest matching degree and the corresponding treatment plan are sent to the livestock user terminal. Otherwise, based on the same principle, the matching degree between the abnormal physical sign manifestations and abnormal behavior manifestations of the abnormal etiology obtained from the second diagnosis and the livestock abnormal physical sign parameters and abnormal behavior images is calculated, and the matching degrees of the abnormal etiologies obtained from the model diagnosis and the second diagnosis are compared. The abnormal etiology and treatment plan corresponding to the higher matching degree value are the finally output abnormal etiology and treatment plan, improving the efficiency and accuracy of disease diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is a system structure block diagram of the present invention.

[0011] Figure 2 It is a method step diagram of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0012] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0013] As Figure 1 shown, this embodiment provides an intelligent diagnosis system for livestock veterinarians based on machine learning, including a physical sign and behavior data acquisition module, a physical sign data preprocessing module, a behavior data preprocessing module, a physical sign abnormality analysis module, a behavior abnormality recognition module, a health intelligent diagnosis module, and a database.

[0014] The physical sign and behavior data acquisition module is connected to the physical sign data preprocessing module and the behavior data preprocessing module. The physical sign data preprocessing module, the physical sign abnormality analysis module, and the health intelligent diagnosis module are sequentially connected. The behavior data preprocessing module, the behavior abnormality recognition module, and the health intelligent diagnosis module are sequentially connected. All modules in the system are connected to the database.

[0015] The physical sign and behavior data acquisition module collects the physical sign data of livestock through multiple sensors configured on the intelligent collar, and obtains the behavior data of livestock through a multi-angle camera; Specifically, in this embodiment, it should be noted that the used intelligent collar integrates a variety of sensors, including a temperature sensor, a heart rate sensor, an acceleration sensor, and a respiratory rate monitoring device. These devices can accurately and real-time monitor and record the physiological data of livestock through high-precision sensors and advanced algorithm technologies. In addition, an electronic tag and a positioning chip are built into the intelligent collar. The electronic tag stores the species name, number, age, and weight of the livestock, and the positioning chip is used to locate the real-time position of the livestock.

[0016] Further, the physical sign behavior data acquisition module includes a physical sign data acquisition unit, a behavior data acquisition unit, and a data transmission unit. The physical sign data acquisition unit collects the body temperature of livestock in real time based on the temperature sensor on the intelligent collar, collects the heart rate of livestock in real time through the heart rate sensor on the intelligent collar, collects the respiratory rate of livestock in real time through the respiratory rate monitoring device on the intelligent collar, and collects the chewing data and movement data of livestock in real time through the acceleration sensor. The livestock chewing data is the Z-axis acceleration collected by the acceleration sensor fixed at the lower jaw of the livestock, and the livestock movement data is the X-axis acceleration and Y-axis acceleration collected by the acceleration sensor fixed on the intelligent collar. The acceleration sensor fixed at the lower jaw of the livestock takes the lower jaw of the livestock as the reference point, and the direction perpendicular to the axis of the head and tail of the livestock and pointing to the side of the livestock body is the Z-axis. The acceleration sensor fixed on the intelligent collar takes the intelligent collar as the reference point. When the livestock moves, the forward direction is the X-axis, and the direction perpendicular to the X-axis and vertically upward is the Y-axis. The behavior data acquisition unit obtains the eating behavior images, movement behavior images, and resting behavior images of livestock through the multi-angle camera. The data transmission unit synchronously transmits the collected livestock physical sign data to the physical sign data preprocessing module and synchronously transmits the obtained livestock behavior images to the behavior data preprocessing module.

[0017] Specifically in this embodiment, it should be noted that body temperature is an important indicator reflecting the metabolic and immune status of livestock. Abnormal body temperature changes usually mean that there are infections, inflammations, or other health problems in the livestock body. By monitoring the body temperature, the fever symptoms of livestock can be detected in time, and it helps veterinarians judge the type and severity of diseases, providing a basis for formulating treatment plans. Movement speed is an important aspect for evaluating the vitality and health status of livestock. Livestock with slow movement or inability to move may be suffering from diseases or other health problems. By monitoring the movement speed, the movement disorders of livestock can be detected in time. Rumination is a unique digestive process of ruminants. A decrease or cessation of rumination time may mean that there are problems in the digestive system of livestock, such as rumen impaction, acidosis, etc. By monitoring the rumination time, the digestive problems of livestock can be detected in time. Heart rate is an important indicator reflecting the health status of the cardiovascular system of livestock. Too fast or too slow heart rate may mean that there are cardiovascular problems or other health problems in the livestock. By monitoring the heart rate, the cardiovascular problems of livestock, such as arrhythmia, heart failure, etc., can be detected in time. Respiratory rate is an important indicator reflecting the health status of the respiratory system of livestock. Abnormal respiratory rate may mean that there are respiratory tract infections, pneumonia, or other respiratory system problems in the livestock. By monitoring the respiratory rate, the respiratory problems of livestock can be detected in time, and the respiratory rate data also helps veterinarians evaluate the ventilation volume and oxygenation ability of livestock, providing a basis for formulating treatment plans. Therefore, it is selected to monitor the real body temperature, movement speed, rumination time, heart rate, and respiratory rate of livestock as the physical sign data for intelligent diagnosis of livestock health to judge whether there are physical sign abnormalities in livestock.

[0018] The vital sign data preprocessing module processes the livestock body surface temperature, livestock chewing data and livestock movement data to obtain the livestock's real body temperature, rumination time, movement state and movement speed; Furthermore, the vital sign data preprocessing module includes a data receiving unit, a body temperature correction unit, a rumination time calculation unit, a motion state analysis unit and a data output unit, wherein the data receiving unit is used to receive the vital sign data of the livestock; the body temperature correction unit inputs the received livestock body surface temperature into the temperature linear regression equation for correction to obtain the real body temperature of the livestock; the rumination time calculation unit uploads the received Z-axis acceleration data to MATLAB, uses the findpeaks function to extract the peak value of the Z-axis acceleration data to obtain the local peak and trough, then sets the positive and negative thresholds to extract the peak data greater than the positive threshold and less than the negative threshold for statistics and mark as rumination, and the proportion of the peak data that meets the requirements in all the collected data is the rumination time of the livestock; the motion state analysis unit takes the received X-axis acceleration and Y-axis acceleration as input, applies the K-means clustering algorithm to divide the livestock motion state into three states of stillness, slow walking and running, and then calculates the real-time motion speed; the data output unit transmits the received livestock heart rate, respiratory rate, real body temperature, rumination time and motion speed to the vital sign abnormality analysis module.

[0019] Specifically, it should be noted that the rectal temperature is usually used as the body temperature of the livestock, and is generally measured manually by a veterinary thermometer. However, this method is labor-intensive, cannot obtain real-time data in a timely manner, and is very likely to cause the spread of diseases. Therefore, an accurate and simple automated temperature measurement method is selected to measure the body temperature of the livestock. The surface temperature of the livestock collected in real time by the temperature sensor of the smart collar will have a certain error with the real body temperature of the livestock. However, since the collected data and the real data are basically linearly related, the collected data can be corrected by linear regression analysis. The predicted body temperature of the livestock obtained after correction is basically consistent with the real body temperature of the livestock, and can be directly used to represent the real body temperature of the livestock.

[0020] Specifically, it should be noted that when constructing the temperature linear regression model, the least squares method can be used to find the best matching function, and the linear regression model is defined as: , y, x, a0, b0 are respectively the true body temperature value, the body surface temperature value, the linear equation coefficient value, and the linear equation error. The expression of the average loss function Le is: , where N is the total number of pairs of body surface temperature and rectal temperature control data of the same livestock collected in the same environment at the same time, i is the i-th control data group, yi is the rectal temperature in the i-th control data group, and xi is the body surface temperature in the i-th control data group. The linear regression model corresponding to the minimum value of the average loss function Le is the best matching model. At this time, the partial derivatives of the average loss function Le with respect to a0 and b0 are both 0, that is: , , after solving, we get: , , retrieve the data in the control data group to simulate and establish a temperature linear regression model.

[0021] Specifically in this embodiment, for the convenience of understanding the calculation rule of rumination time, a set of data is given for demonstration. Assume that the Z-axis accelerations after peak extraction are a1 = 10m / s 2 , a2 = 3m / s 2 , a3 = -1m / s 2 , a4 = -5m / s 2 , a5 = 2.5m / s 2 , a6 = -3m / s 2 , a7 = 6m / s 2 , a8 = 4.5m / s 2 There are a total of five groups of peak data and three groups of trough data. The positive and negative of the acceleration only represent the direction. The set thresholds are 5m / s 2 and -5m / s 2 , then there are a total of two groups of peak data with accelerations greater than 5m / s 2 and less than -5m / s 2 . The proportion in all the collected data is 25%. The collection time is 1h, so the rumination time is 15min.

[0022] The behavior data preprocessing module uses an image processing tool to extract the feeding posture image, feeding action image, movement posture image, and lying posture image of the livestock from the behavior data and identify whether the extracted posture images are accurate, and updates the behavior data based on the recognition results; Further, the behavior data preprocessing module includes a data receiving unit, a behavior extraction unit, an image recognition unit, a behavior data recording unit, and a data output unit. The data receiving unit is used to receive the obtained livestock behavior images; the behavior extraction unit uses image processing tools to extract the feeding postures and feeding actions of livestock from the diet behavior image videos, extracts the movement postures of livestock from the movement behavior image videos, and extracts the lying postures of livestock from the rest behavior images; the image recognition unit identifies whether the extracted posture images are correct based on image recognition algorithms; the behavior data recording unit is used to record the cumulative duration of diet behavior, the cumulative duration of movement behavior, the cumulative duration of rest behavior, the duration of correct feeding postures, the duration of correct feeding actions, the duration of correct movement postures, and the duration of correct lying postures; the data output unit transmits the recorded real-time cumulative duration of diet behavior, the cumulative duration of movement behavior, the cumulative duration of rest behavior, the cumulative duration of correct feeding postures, the cumulative duration of correct feeding actions, the cumulative duration of correct movement postures, and the cumulative duration of correct lying postures to the behavior anomaly recognition module.

[0023] Specifically in this embodiment, MultimodalVideoTag or VideoTag can be applied to classify and extract livestock behavior data. The former is based on real short video service data, integrates three modalities of video text, image, and audio for video multi-modal tag classification. The model provides 25 first-level tags and more than 200 second-level tags, and the tag accuracy rate exceeds 85%; the latter is based on tens of millions of data of Baidu's short video service, supports 3,000 practical tags derived from industrial practices, has good generalization ability, and is very suitable for applications in large-scale (tens of millions, hundreds of millions, billions) short video classification scenarios in China, and the tag accuracy rate reaches 89%.

[0024] Specifically in this embodiment, the eating posture and movements can reflect the oral cavity, digestive system and overall health status of livestock. For example, abnormal eating postures may indicate oral pain, digestive system diseases or other health problems in livestock; the movement posture can reveal the status of the muscles, bones, nervous system and overall activity ability of livestock, and abnormal movements may indicate diseases, pain or movement disorders; for some livestock (such as dairy cows), the lying posture is an important indicator for evaluating their comfort, rest quality and potential health problems. For example, a decrease in the lying time of dairy cows may be related to heat stress, pain or production stress in them. These behavioral data can usually be easily obtained through video monitoring or on-site observation without complex equipment or technology. The recording and analysis of behavioral data are relatively intuitive, facilitating veterinarians or breeders to quickly identify potential health problems. Compared with traditional diagnostic methods such as blood tests and tissue biopsies, intelligent diagnosis based on behavioral data has the advantage of being non-invasive, reducing the stress response and potential harm to livestock. By continuously monitoring and analyzing these behavioral data, potential diseases or health problems can be detected before livestock show clinical symptoms, enabling timely intervention measures to be taken. Combining machine learning and artificial intelligence technologies, these behavioral data can be deeply analyzed and pattern recognized to improve the accuracy and reliability of diagnosis. Therefore, choosing the eating posture, eating movements, movement posture and lying posture as the behavioral data for intelligent diagnosis of livestock health has the advantages of reflecting the health status, being easy to observe and record, and being non-invasive.

[0025] The physical sign abnormality analysis module inputs the livestock heart rate, respiratory rate, true body temperature, rumination time and movement speed into the livestock physical sign abnormality recognition model that has been trained and passed the test for abnormality recognition, marks the physical sign abnormal data group, and sends the livestock physical sign abnormal data group to the health intelligent diagnosis module after confirming the physical sign abnormality of the livestock; Furthermore, after receiving the livestock heart rate, respiratory rate, true body temperature, rumination time and movement speed, the physical sign abnormality analysis module inputs the received physical sign data into the livestock physical sign abnormality recognition model that has been trained and passed the test for abnormality recognition, outputs the abnormality determination results, abnormality coefficient and physical sign comprehensive abnormality index of each physical sign parameter. The abnormality coefficient βxi of the i-th physical sign parameter is determined by the positive and negative relationship between the actual value βci of the received i-th physical sign parameter and the upper limit value βsai and lower limit value βsbi of the corresponding latest normal physical sign parameter in the model. The specific formula is: , and the specific calculation formula for the physical sign comprehensive abnormality index αz is: , \(mzy\) and \(mzx\) are the numbers of abnormal physical sign parameters and normal physical sign parameters respectively. Subsequently, the calculated comprehensive physical sign abnormality index is compared with the preset value of physical sign abnormality. If the calculated value is greater than the set value, the physical sign data of this group of livestock is marked as an abnormal data group. If abnormal data groups exceeding the preset number of groups continuously appear, it is determined that the physical signs of the livestock are abnormal, and the abnormal physical sign data group of the livestock is sent to the health intelligent diagnosis module.

[0026] Specifically, it should be noted in this embodiment that the training data set and validation data set of the livestock physical sign abnormality recognition model used are the normal physical sign data and abnormal physical sign data of the whole life cycle of several groups of livestock. The specific model construction process is prior art and will not be elaborated here.

[0027] Specifically, it should be noted in this embodiment that after the model training is completed, the normal standards of the physical sign parameters of livestock corresponding to the age, weight and environmental conditions will be generated based on the training data. When the received heart rate, respiratory rate, real body temperature, rumination time and movement speed of the livestock are input into the trained and tested livestock physical sign abnormality recognition model for abnormality recognition, the model will automatically compare the physical sign parameters of the received data with the normal standards of the physical sign parameters of livestock corresponding to the age, weight and environmental conditions. If the actual physical sign parameters meet the standards, the name of the physical sign parameter will be output, followed by the index normal and the abnormality coefficient with a value of 0. If the actual physical sign parameters do not meet the standards, the name of the physical sign parameter will be output, followed by the index on the high side or the index on the low side and the specific value of the abnormality coefficient.

[0028] Specifically in this embodiment, for the convenience of understanding the process scheme of abnormal sign analysis, a set of examples are provided: A calf aged one year and weighing 116 kg has a heart rate, respiratory rate, true body temperature, rumination time, and slow walking speed of 70 beats / minute, 40 breaths / minute, 32.7 °C, 15 minutes, and 50 m / minute respectively after a set of treatments in summer. The normal heart rate range, normal respiratory rate range, normal true body temperature range, normal rumination time range, and normal movement speed range of a calf aged one year and weighing 100 - 120 kg generated by the model in summer are [60 beats / minute, 75 beats / minute], [45 breaths / minute, 55 breaths / minute], [31.5 °C, 32.5 °C], [20 minutes, 45 minutes], and [10 m / minute, 100 m / minute] respectively. Then the corresponding abnormal recognition results are: the heart rate index is normal, and the abnormal degree coefficient of the heart rate is 0; the respiratory rate index is low, and the abnormal degree coefficient of the respiratory rate is (45 - 40) / 45 = 0.1111; the true body temperature index is high, and the abnormal degree coefficient of the true body temperature is (32.7 - 32.5) / 32.5 = 0.0062; the rumination time index is low, and the abnormal degree coefficient of the rumination time is (20 - 15) / 20 = 0.2500; the slow walking speed index is normal, and the abnormal degree coefficient of the slow walking speed is 0. Then the comprehensive abnormal degree index of the calf's signs is 4 / 3×(0.0062 + 0.2500 + 0 + 0 + 0.1111) = 0.4897. Then, the calculated comprehensive abnormal degree index of the calf's signs is compared with the preset value. If the calculated value is greater than the preset value, the actual sign parameters of this group of calves are marked as an abnormal data group. Otherwise, only the abnormal sign parameters of the calves are marked. If there are continuously abnormal data groups exceeding the preset number of groups, it is determined that the livestock signs are abnormal, and the abnormal livestock sign data group is sent to the health intelligent diagnosis module. If the abnormal preset value is 0, the example data group of the calves given is the abnormal data group. If the preset number of groups is 5, assuming that the comprehensive abnormal degree indexes of the signs of 5 groups of calves are 0.4987, 0.2573, 0.1986, 0.2471, and 0.3976 in sequence, then these five groups of sign parameters are all abnormal data groups and trigger the determination condition of calf sign abnormality, and directly output the calf sign abnormality and the 5 groups of abnormal sign data of the calves to the health intelligent diagnosis module.

[0029] The behavior abnormal recognition module calculates the proportion of correct postures of eating behavior, exercise behavior, and rest behavior, then calculates the abnormal degree coefficient of each behavior, and then calculates the comprehensive abnormal degree index of behavior, marks the abnormal image data group, and sends the abnormal livestock behavior image data group to the health intelligent diagnosis module after confirming the livestock behavior abnormality; Furthermore, after receiving the cumulative duration of real-time eating behavior, cumulative duration of exercise behavior, cumulative duration of rest behavior, cumulative duration of correct eating postures, cumulative duration of correct eating actions, cumulative duration of correct exercise postures, and cumulative duration of correct lying postures, the behavior anomaly recognition module calculates the anomaly coefficient and the comprehensive behavior anomaly index. The anomaly coefficient βyi of the i-th behavior type is determined by the positive or negative relationship between the proportion βzi of correct postures of the i-th behavior type received and the set proportion βri of correct postures. The specific formula is: , the proportion of correct postures of eating behavior is the product of the sum of the cumulative duration of correct eating postures and the cumulative duration of correct eating actions and the cumulative duration of eating behavior. The proportion of correct postures of exercise behavior is the product of the cumulative duration of correct exercise postures and the cumulative duration of exercise behavior. The proportion of correct postures of rest behavior is the cumulative duration of correct lying postures and the cumulative duration of rest behavior. The specific calculation formula of the comprehensive behavior anomaly index αx is: , mxa and mxb are the number of abnormal behavior types and the number of normal behavior types in turn. Then, the calculated comprehensive behavior anomaly index is compared with the preset value of behavior anomaly. If the calculated value is greater than the set value, the error posture images of various existing behaviors are extracted and marked as an abnormal image data group. If the marking operation exceeds the preset number of times continuously, it is determined that the livestock behavior is abnormal, and the livestock behavior abnormal image data group is sent to the health intelligent diagnosis module.

[0030] The health intelligent diagnosis module inputs the received livestock physical sign abnormal data group and livestock behavior abnormal image data group into the trained and tested livestock health intelligent diagnosis model for health diagnosis, outputs the abnormal cause with the highest matching degree and the corresponding treatment plan, determines whether the model diagnosis result triggers a secondary diagnosis. If it triggers, a secondary diagnosis is performed and the final output abnormal cause and treatment plan are determined based on the set rules. If it does not trigger, the abnormal cause and treatment plan output by the model diagnosis are directly output.

[0031] Further, after receiving the livestock physical sign abnormal data group and the livestock behavior abnormal image data group, the health intelligent diagnosis module inputs the received livestock physical sign abnormal data group and the livestock behavior abnormal image data group as the input end into the livestock health intelligent diagnosis model that has been trained and passed the test for health diagnosis. The model automatically matches the physical sign abnormal parameters and the behavior abnormal images with the abnormal physical sign manifestations and abnormal behavior manifestations corresponding to different etiologies in the machine learning process one by one. The ratio of the number of overlapping abnormal parameters to the number of actual abnormal parameters of the livestock is the matching degree. The abnormal etiology with the highest matching degree and the corresponding treatment plan are output as the output end. Then, the diagnosis result is determined, that is, the abnormal etiology matching degree is compared with the set matching degree confidence threshold. If the abnormal etiology matching degree is greater than or equal to the set matching degree confidence threshold, the abnormal etiology with the highest matching degree and the corresponding treatment plan can be directly sent to the livestock user terminal. Otherwise, information is sent to multiple veterinarians registered in the system to request a second diagnosis. If the second diagnosis result is consistent with the model diagnosis result, the abnormal etiology with the highest matching degree and the corresponding treatment plan are sent to the livestock user terminal. Otherwise, based on the same principle, the matching degree between the abnormal physical sign manifestations and abnormal behavior manifestations of the abnormal etiology obtained from the second diagnosis and the livestock abnormal physical sign parameters and abnormal behavior images is calculated, and the matching degrees of the abnormal etiologies obtained from the model diagnosis and the second diagnosis are compared. The abnormal etiology and treatment plan corresponding to the one with a higher matching degree value are the final output abnormal etiology and treatment plan.

[0032] Specifically, it should be noted in this embodiment that the learning data and verification data of the used livestock health intelligent diagnosis model are the existing livestock disease types and their abnormal manifestation data.

[0033] The database is used to store the data information of all modules in the system.

[0034] Specifically, it should be noted in this embodiment that the preset values or standard values used are selected based on actual needs or data rules, and no specific value selection is limited here.

[0035] As Figure 2 shown, this embodiment provides a livestock veterinary intelligent diagnosis method based on machine learning, which specifically includes the following steps: S1: Collect the physical sign data of livestock through multiple sensors configured on the intelligent collar, and obtain the behavior data of livestock through multi-angle cameras; S2: Process the body surface temperature, chewing data, and movement data of livestock to obtain the true body temperature, rumination time, movement state, and movement speed of livestock; S3: Use image processing tools to extract the feeding posture image, feeding action image, movement posture image, and lying posture image of livestock from the behavior data and identify whether the extracted posture images are accurate. Update the behavior data based on the recognition results; S4: Input the livestock heart rate, respiratory rate, real body temperature, rumination time, and movement speed into the livestock physical sign abnormality recognition model that has been trained and passed the test for abnormality recognition, mark the physical sign abnormal data group, and output the physical sign abnormal data group after confirming the livestock physical sign abnormality; S5: Calculate the proportion of correct postures of eating behavior, movement behavior, and rest behavior, then calculate the abnormality coefficient of each behavior, and then calculate the comprehensive behavior abnormality index, mark the abnormal image data group, and output the behavior abnormal image data group after confirming the livestock behavior abnormality; S6: Use the received livestock physical sign abnormal data group and livestock behavior abnormal image data group as the input end and input them into the livestock health intelligent diagnosis model that has been trained and passed the test for health diagnosis, output the abnormal cause of illness with the highest matching degree and the corresponding treatment plan, determine whether the model diagnosis result triggers a secondary diagnosis. If it triggers, conduct a secondary diagnosis and determine the abnormal cause of illness and treatment plan finally output based on the set rules. If it does not trigger, directly output the abnormal cause of illness and treatment plan output by the model diagnosis.

[0036] Finally: The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An intelligent diagnosis system for animal husbandry and veterinary medicine based on machine learning, characterized in that: include: Vital sign and behavior data collection module: collects livestock's vital sign data through multiple sensors configured on the smart collar, and obtains livestock's behavior data through multi-angle cameras; Vital sign data preprocessing module: processes the livestock surface temperature, livestock chewing data and livestock movement data to obtain the livestock's real body temperature, rumination time, movement state and movement speed; Behavior data preprocessing module: using image processing tools to extract livestock's eating posture images, eating action images, movement posture images, and lying posture images from the behavior data and identify whether the extracted posture images are accurate, and updating the behavior data based on the identification results; Abnormal signs analysis module: input the livestock's heart rate, respiratory rate, real body temperature, rumination time and movement speed into the livestock abnormal signs recognition model that has been trained and tested to perform abnormality recognition, mark the abnormal signs data group, and send the abnormal signs data group to the health intelligent diagnosis module after confirming that the livestock's physical signs are abnormal; Abnormal behavior recognition module: calculates the percentage of correct postures for eating, exercising, and resting behaviors, then calculates the abnormality coefficient of each behavior, and then calculates the comprehensive abnormality index of the behavior, marks the abnormal image data group, and sends the abnormal livestock behavior image data group to the health intelligent diagnosis module after confirming that the livestock behavior is abnormal; Health intelligent diagnosis module: The received livestock physical sign abnormality data group and livestock behavior abnormality image data group are input as input into the trained and tested livestock health intelligent diagnosis model for health diagnosis, and the abnormal cause and corresponding treatment plan with the highest matching degree are output. It is determined whether the model diagnosis result triggers a secondary diagnosis. If it is triggered, a secondary diagnosis is performed and the final output abnormal cause and treatment plan are determined based on the set rules. If it is not triggered, the abnormal cause and treatment plan output by the model diagnosis are directly output.

2. The animal husbandry and veterinary intelligent diagnosis system based on machine learning according to claim 1, characterized in that: The vital sign behavior data acquisition module includes a vital sign data acquisition unit, a behavior data acquisition unit and a data transmission unit. The vital sign data acquisition unit collects the body surface temperature of the livestock in real time based on the temperature sensor on the smart collar, collects the heart rate of the livestock in real time through the heart rate sensor on the smart collar, collects the respiratory rate of the livestock in real time through the respiratory rate monitoring device on the smart collar, and collects the chewing data and livestock movement data of the livestock in real time through the acceleration sensor. The livestock chewing data is the Z-axis acceleration collected by the acceleration sensor fixed at the lower jaw of the livestock, and the livestock movement data is the X-axis acceleration and the Y-axis acceleration collected by the acceleration sensor fixed on the smart collar. The acceleration sensor fixed at the lower jaw of the livestock takes the lower jaw of the livestock as a reference point, and the direction perpendicular to the axis of the head and tail of the livestock and pointing to the side of the livestock is the Z axis. The acceleration sensor fixed on the smart collar takes the smart collar as a reference point, and the direction of the livestock moving forward is the X axis, and the direction perpendicular to the X axis and vertically upward is the Y axis. The behavior data acquisition unit obtains the eating behavior image, exercise behavior image and resting behavior image of the livestock through a multi-angle camera; The data transmission unit synchronously transmits the collected livestock vital sign data to the vital sign data preprocessing module, and synchronously transmits the acquired livestock behavior images to the behavior data preprocessing module.

3. The animal husbandry and veterinary intelligent diagnosis system based on machine learning according to claim 1, characterized in that: The vital sign data preprocessing module includes a data receiving unit, a body temperature correction unit, a rumination time calculation unit, a motion state analysis unit and a data output unit, and the data receiving unit is used to receive the livestock vital sign data; The body temperature correction unit inputs the received livestock body surface temperature into the temperature linear regression equation for correction to obtain the livestock's real body temperature; The rumination time calculation unit uploads the received Z-axis acceleration data to MATLAB, uses the findpeaks function to extract the peak value of the Z-axis acceleration data to obtain the local peaks and troughs, and then sets the positive and negative thresholds to extract the peak data greater than the positive threshold and less than the negative threshold for statistics and mark them as rumination. The proportion of peak data that meets the requirements in all the collected data is the rumination time of the livestock; The motion state analysis unit takes the received X-axis acceleration and Y-axis acceleration as input, applies the K-means clustering algorithm to classify the livestock motion state into three states: still, slow walking and running, and then calculates the real-time motion speed; The data output unit transmits the received livestock heart rate, respiratory rate, real body temperature, rumination time and movement speed to the abnormal vital signs analysis module.

4. The animal husbandry and veterinary intelligent diagnosis system based on machine learning according to claim 1, characterized in that: The behavior data preprocessing module includes a data receiving unit, a behavior extraction unit, an image recognition unit, a behavior data recording unit and a data output unit, wherein the data receiving unit is used to receive the acquired livestock behavior images; the behavior extraction unit uses an image processing tool to extract the livestock's eating posture and eating action from the eating behavior image video, extracts the livestock's movement posture from the movement behavior image video, and extracts the livestock's lying posture from the resting behavior image; the image recognition unit identifies whether the extracted posture image is correct based on an image recognition algorithm; the behavior data recording unit is used to record the cumulative duration of eating behavior, the cumulative duration of movement behavior, the cumulative duration of resting behavior, the duration of correct eating posture, the duration of correct eating action, the duration of correct movement posture and the duration of correct lying posture; the data output unit transmits the recorded real-time cumulative duration of eating behavior, cumulative duration of movement behavior, cumulative duration of resting behavior, cumulative duration of correct eating posture, cumulative duration of correct eating action, cumulative duration of correct movement posture and cumulative duration of correct lying posture to the behavior abnormality recognition module.

5. The animal husbandry and veterinary intelligent diagnosis system based on machine learning according to claim 1, characterized in that: After receiving the heart rate, respiratory rate, real body temperature, rumination time and movement speed of the livestock, the abnormal sign analysis module inputs the received physical sign data into the trained and tested livestock physical sign abnormality recognition model for abnormal recognition, and outputs the abnormality determination results and abnormality coefficients of various physical sign parameters and the physical sign comprehensive abnormality index. The abnormality coefficient βxi of the i-th physical sign parameter is determined by the positive and negative relationship between the actual value βci of the i-th physical sign parameter received and the upper and lower limits βsai and βsbi of the latest normal physical sign parameter fluctuations corresponding to the model. The specific formula is: , the specific calculation formula of the comprehensive abnormality index αz of physical signs is: , mzy and mzx are the number of abnormal physical sign parameters and the number of normal physical sign parameters respectively. The calculated physical sign comprehensive abnormality index is then compared with the preset value of physical sign abnormality. If the calculated value is greater than the set value, the group of livestock physical sign data is marked as an abnormal data group. If abnormal data groups exceeding the preset number of groups appear continuously, the livestock physical signs are judged to be abnormal, and the abnormal livestock physical sign data group is sent to the health intelligent diagnosis module.

6. The animal husbandry and veterinary intelligent diagnosis system based on machine learning according to claim 1, characterized in that: The abnormal behavior recognition module receives the real-time cumulative duration of eating behavior, cumulative duration of exercise behavior, cumulative duration of rest behavior, cumulative duration of correct eating posture, cumulative duration of correct eating action, cumulative duration of correct exercise posture and cumulative duration of correct lying posture, and calculates the abnormality coefficient and the comprehensive abnormality index of behavior. The abnormality coefficient βyi of the i-th behavior type is determined by the positive and negative relationship between the difference between the received correct posture proportion βzi of the i-th behavior type and the set correct posture proportion βri. The specific formula is: The correct posture ratio of eating behavior is the product of the cumulative duration of correct eating posture, the cumulative duration of correct eating movements and the cumulative duration of eating behavior. The correct posture ratio of exercise behavior is the product of the cumulative duration of correct exercise posture and the cumulative duration of exercise behavior. The correct posture ratio of resting behavior is the cumulative duration of correct lying posture and the cumulative duration of resting behavior. The specific calculation formula of the comprehensive abnormal behavior index αx is: , mxa and mxb are the number of abnormal behavior types and the number of normal behavior types respectively. The calculated comprehensive abnormality index of behavior is then compared with the preset value of abnormal behavior. If the calculated value is greater than the set value, the existing incorrect posture images of various behaviors are extracted and marked as abnormal image data groups. If the marking operation occurs continuously for more than the preset number of times, the livestock behavior is judged to be abnormal, and the abnormal livestock behavior image data group is sent to the health intelligent diagnosis module.

7. The animal husbandry and veterinary intelligent diagnosis system based on machine learning according to claim 1, characterized in that: After receiving the livestock abnormal physical sign data group and the livestock abnormal behavior image data group, the health intelligent diagnosis module inputs the received livestock abnormal physical sign data group and the livestock abnormal behavior image data group as input ends into the livestock health intelligent diagnosis model that has been trained and tested for health diagnosis. The model automatically matches the abnormal physical sign parameters and abnormal behavior images with the abnormal physical sign manifestations and abnormal behavior manifestations corresponding to different causes in the machine learning process one by one. The ratio of the number of overlapping abnormal parameters to the actual number of abnormal parameters of the livestock is the matching degree. The abnormal cause with the highest matching degree and the corresponding treatment plan are output as output ends, and then the diagnosis result is judged, that is, the abnormal cause matching degree is compared with the set matching degree credible threshold. If the matching degree of the abnormal cause is greater than or equal to the set matching degree trust threshold, the abnormal cause with the highest matching degree and the corresponding treatment plan can be directly sent to the animal husbandry user end. Otherwise, information is sent to multiple veterinarians registered in the system to request a secondary diagnosis. If the secondary diagnosis result is consistent with the model diagnosis result, the abnormal cause with the highest matching degree and the corresponding treatment plan are sent to the animal husbandry user end. Otherwise, based on the same principle, the matching degree of the abnormal signs and abnormal behavior manifestations of the abnormal cause obtained by the secondary diagnosis with the abnormal signs parameters and abnormal behavior images of livestock is calculated, and the matching degree of the abnormal cause obtained by the model diagnosis and the secondary diagnosis is compared. The abnormal cause and treatment plan corresponding to the one with a higher matching degree value are the abnormal cause and treatment plan finally output.

Images