TWI766803B - Livestock diet monitoring method and system - Google Patents

Abstract

The present invention is a method and system for monitoring the diet of livestock, which can detect the health status of livestock at an early stage by monitoring the diet behavior of livestock. The system includes a label device, a detection device and a processing device, wherein the processing device further includes a storage module, a collection module, an analysis module and a classification module. The present invention uses the detection device to collect label data of the label device, and convert the label data Imported into the processing device, and calculate the daily eating time and the total number of meals of livestock. In addition, the present invention uses the processing device to extract four features, such as state feature, round feature, continuation feature and meal interval feature, and then use the above features. As a variable, it is substituted into the probability function to obtain the similarity rate of eating behavior, so as to achieve the purpose of monitoring the daily eating time and total eating frequency of livestock with a simple detection device.

Description

Livestock diet monitoring method and system

The present invention relates to a livestock diet monitoring method and system, in particular, to a method and system for recording livestock behaviors by electronic tags to obtain livestock behavioral characteristics, and using the livestock behavioral characteristics as variables to substitute into a probability function (P) based on Bayes' theorem, and borrowing The dietary status of livestock can be inferred from the probability ratio between diet and non-diet, also known as the similarity rate of eating behavior, so as to achieve the purpose of diet monitoring.

The need for livestock farmers to grasp the health status of livestock is as important to the livelihood of livestock farmers as it is for manufacturers to grasp the production status. Traditionally, animal husbandry people used to manually record the sleep, diet, activity, and other living behaviors of livestock as indicators of health status. Among them, eating behavior is one of the most important indicators. Animal husbandry can infer the physical condition of livestock through eating behavior. As a result, many prior techniques have been developed to accurately grasp the diet of livestock.

Chinese patent CN103488148B discloses an intelligent monitoring system for livestock behavior, which uses a plurality of motion sensors, geomagnetic sensors, sound sensors, and temperature sensors installed in a breeding environment to sense the diet and excretion of livestock. , rest and other life behaviors. There is another Chinese patent CN107182909A, which discloses a method for feeding pigs. The weight scale, thermometer and hygrometer installed in the breeding environment are used to detect the weight and environmental conditions of livestock, and adjust its parameters according to the variety of livestock, infer that the livestock is in the breeding environment. Eating behavior during growth stages. However, the above-mentioned prior techniques are mainly used for post-hoc analysis, and the present technique can be used for the purpose of real-time monitoring of the daily eating time, total number of eating times and eating behavior of livestock.

In addition, there is a prior art that uses monitoring equipment and an image recognition tracking system to record the long-term activity behavior of livestock, and determine the eating behavior of livestock based on whether the livestock is close to the feed area. However, when the livestock is too concentrated or the light is dim, the recognition ability of the image recognition tracking system is easily affected, so that the behavior of individual livestock cannot be accurately identified.

In order to solve the problem of the above image recognition tracking system, there is a research paper (Adrion Flex et.al.(2018).Monitoring trough visits of growing-finishing pigs with UHF-RFID. Computers and Electronics in Agriculture ,vol.144,pp.144 -153.), replacing the image recognition tracking system with an electronic label system. However, the feeding behavior of livestock is still determined according to whether the livestock is close to the feed area, but the livestock may not be eating when they are close to the feed area. Therefore, it is easy to cause misjudgment of eating behavior by only taking the proximity of the feed area as the judgment criterion.

To sum up, the prior art has the following problems:

1. The accuracy of the multi-sensing system is limited by the detection range and environmental changes, and it is not easy to commercialize.

2. The image recognition tracking system has poor recognition ability when livestock are too concentrated or the light is dim, and cannot accurately identify the behavior of individual pigs.

3. It is easy to cause misjudgment only by judging the eating behavior of livestock by being close to the feed area.

In view of the above problems, the purpose of the present invention is to provide a method and system for monitoring the diet of livestock. The proposed calculation method can calculate the daily eating time and the total number of meals of livestock. In addition, the processing device extracts state features by analyzing the tag data, returns Four characteristics, such as combined characteristics, continuation characteristics, and meal-separation characteristics, are used as variables, and a probability function (P) is substituted into the above four characteristics to infer the possibility of the subsequent eating behavior of livestock, so as to achieve the goal of monitoring the subsequent eating behavior of livestock. Purpose.

The invention provides a livestock diet monitoring method, which is applied to the breeding space of livestock, wherein the livestock diet monitoring method comprises placing a label device representing the livestock on a target, and using a detection device provided on a feed container to monitor the feed at regular intervals. Detect a label device within its detection range once, and obtain label data, wherein each label data includes the number of the label device, the signal strength, and the detected time.

A filtering unit is used to compare each tag data by a tree structure filtering method, and the duplicate tag data is removed. Then, the number of detections of each label device is calculated, and the label data whose number of consecutive detections is too small is removed, so as to achieve the purpose of rapid comparison.

Using a sorting unit, a data set corresponding to the serial number of the labeling device is generated, and the label data with the same serial number is imported into the same data set in chronological order.

Use a state unit to form an event group with consecutive tag data in the data set, and calculate the signals between all tag data in the event group to which it belongs and before the time point when the tag data is detected The intensity is poor, and finally the state characteristics are judged as diet or non-diet based on all the label data in the event group.

At the same time, check whether the tag data is the first data in the event group. If it is the first data, its signal strength is temporarily stored, and the number of data in the event group is +1. If the signal strength difference of the tag data is greater than the maximum signal If the intensity difference (ΔS ₀ ), and the time interval of the continuously read data is less than the minimum duration (T ₀ ), the state characteristic of the animal is judged as eating, otherwise it is not eating.

Using a round unit, calculate the round time difference between each label data whose state feature is diet and the previous label data whose status feature is food in each data set, and assign each label data a round feature, where the round time difference is less than the interval The round characteristics of each label data of a round (Bout Criterion Interval, BCI) are judged as being in the round, otherwise it is out of the round; Same meal.

A continuum unit is used to assign a continuum feature to each tag data. If the state feature of each tag data is the same as that of the previous tag data, the calculated state features are consecutively the same number of times, and the continuum feature is the same continuous state. number of times. Wherein, if the state feature of the label data is different from that of the previous label data, the continuous state feature is reset to the initial value 1; Meal time for a meal.

A meal interval unit is used to assign a meal interval feature to each label data. If each label data meets the precondition of a meal interval, the meal interval time difference of each label data is calculated, and the meal interval time difference is greater than the meal interval time (Time Between Meal Intervals). Meals, TBM), the meal interval feature of each label data is set as long-term, otherwise it is short-term. Wherein, if each label data does not meet the precondition for meal separation, the meal separation feature of each label data is set as short-term, and the time when each label data is detected is set as the initial time.

Among them, the preconditions for meal interval are that the state feature of the label data is diet, the round feature is the same, and the continuous state feature is greater than the continuous threshold (Continuous Threshold, CT).

A computing unit is used to evaluate the target's total eating time and total eating times according to the round time difference and round characteristics.

A behavior identification unit is used to calculate the similarity rate of livestock eating behavior, and if the similarity rate is greater than 1, it is judged that the livestock will eat. If the similarity rate is less than 1, it is judged that the livestock will not be in the feeding state; otherwise, it is judged that the animal is in the feeding state. The similarity ratio is calculated by using a probability function (P), as shown in the following formula (1):

其中，L _i 代表目標的第i筆標籤資料的續態特徵、D _i 表示目標的第i筆標籤資料的回合特徵、f _i 表示目標的第i筆標籤資料的隔餐特徵、S _i 表示目標的第i筆標籤資料的狀態特徵、S _i+1=1表示目標的第i+1筆標籤資料的狀態特徵為飲食、S _i+1=0表示目標的第i+1筆標籤資料的狀態特徵為非飲食，P(x)表示機率函數。

Wherein, Li represents the continuation feature of the ith tag data of the target, D _i represents the round feature of the ith tag data of the target, f _i represents the interval feature of the ith tag data of the target, and S _i represents _the target The state feature of the i -th tag data, S _{i +1} =1 indicates that the state feature of the i+1- th tag data of the target is diet, and S i ₊₁ =0 indicates the state of the i-th +1-th tag data of the target The feature is non-diet, and P(x) represents the probability function.

The invention provides a livestock diet monitoring system, which is applied to a breeding space, wherein the breeding space is provided with a feed container, and the livestock as a monitored target, wherein the livestock diet monitoring system includes: a label device, a detection device, and a processing device and display device.

The detection device is installed on the feed container. When the target with the label device is close to the feed container, the detection device detects the label device and obtains the label data such as the number, signal strength, and detected time of the label device.

The processing device includes a storage module, a collection module, an analysis module and a classification module. The storage module includes a temporary storage unit and a plurality of data sets, the collection module includes a filter unit, a classification unit and a status unit, the analysis module includes a calculation unit and a round unit, and the classification module is a Bayesian classifier. The classification module includes a behavior recognition unit, and the classification module also includes a feature sub-module, wherein the feature sub-module includes a continuation unit and a meal separation unit.

Among them, the classification unit is used to import the data of each tag with the same tag number into the same data set in chronological order.

Wherein, the state unit analyzes the constituent event according to the label data, and judges that the state characteristic of the event is eating or non-eating.

The round unit is used to calculate the round time interval of each tag data, and find out the tag data of the same round.

Wherein, the continuous state unit is used to calculate the state characteristics of the label data for the same number of consecutive times as the continuous state characteristics of the label data.

Among them, the meal interval unit is used to calculate the meal interval difference between the time when each label data that meets the precondition for meal interval is detected and its nearest initial time, and the difference between meal intervals is greater than the interval time between meals (Time between meals). , TBM), the meal interval feature of the label data is set to long-term.

Among them, the behavior recognition unit is used to use the feature sub-module to determine the similarity rate between the latest label data and past features. If the similarity rate is greater than 1, it is determined that the target livestock will eat, and the target whose similarity rate is less than 1 will not eat. .

1: Processing device

10: Storage Module

101: Datasets

102: Temporary storage unit

11: Collect Mods

111: Filter unit

112: Classification Unit

113: Status Unit

12: Analysis module

121: Round Unit

122: Computing Unit

13: Classification module

130: Feature Sub-Module

131: Continued state unit

132: Meal Separation Unit

133: Behavior Recognition Unit

2: Label device

3: Detection device

30: Label Information

4: Livestock

5: Feed container

6: Breeding space

7: Display device

S301-S309: Steps

S401-S407: Steps

S701-S708: Steps

S801-S803: Steps

Fig. 1 is the usage situation diagram of the livestock diet monitoring system of the present invention;

Fig. 2 is a system architecture diagram of the livestock diet monitoring system of the present invention;

Fig. 3 is the flow chart of the livestock diet monitoring method of the present invention;

Fig. 4 is the filtering flow chart of the livestock diet monitoring method of the present invention;

5A and 5B are schematic diagrams of the tree structure filtering method of the livestock diet monitoring method of the present invention;

FIG. 6 is a schematic diagram of an embodiment of label data classification according to the present invention;

Fig. 7 is the state judgment flow chart of the livestock diet monitoring method of the present invention;

FIG. 8 is a flowchart of the present invention for judging the characteristics of meal intervals;

In order to facilitate the examiners to understand the features, contents and advantages of the present invention and the effects that can be achieved, the present invention is hereby described in detail with the accompanying drawings and in the form of embodiments as follows, and the drawings used in the text, Its purpose is only for illustration and auxiliary description, so it should not limit the patent scope of the present invention in actual implementation.

Please refer to FIG. 1 , which is a usage scenario diagram of the livestock diet monitoring system of the present invention. In one embodiment of the present invention, the livestock diet monitoring system of the present invention includes a processing device 1 , a labeling device 2 , a detection device 3 and a display device 7 , wherein the labeling device 2 and the detection device 3 are installed in the breeding of the captive livestock 4 In the space 6, and the label device 2 is arranged on the body of each livestock 4 to represent each livestock 4, wherein the detection device 3 is arranged above the feed container 5, when the livestock 4 is close to the feed container 5, the detection device 3 The tag device 2 within its detection range is detected to obtain tag data 30 , and the tag data 30 is wirelessly transmitted to the processing device 1 .

Please refer to FIG. 2 , which is a system architecture diagram of the livestock diet monitoring system of the present invention. The processing device 1 of the livestock diet monitoring system of the present invention further includes a storage module 10, a collection module 11, an analysis module 12 and a classification module 13, wherein the collection module 11 is connected to the analysis module 12, and the classification module The group 13 is connected to the analysis module 12 , and the storage module 10 is simultaneously connected to the collection module 11 , the analysis module 12 and the classification module 13 .

The collection module 11 further includes a filter unit 111, a classification unit 112 and a status unit 113, wherein the filter unit 111 reads the tag data 30 from the detection device 3, filters the tag data 30, and finally uploads the tag data 30 To the storage module 10, the categorization unit 112 reads the storage module 10, and categorizes each tag data 30 into different data sets 101, and then the status unit 113 determines whether the eating behavior of each tag data 30 is diet or non-food. Eating and drinking, and setting the eating behavior of each tag data 30 as a state feature of each tag data 30 .

The analysis module 12 further includes a round unit 121 and a calculation unit 122, wherein the round unit 121 further analyzes each label data 30 whose state feature is diet, and determines whether each label data 30 and the previous label data 30 belong to the same round, and give each tag data 30 corresponding round characteristics. Specifically, between the previous tag data 30 is the same meal between each tag data 30 in the round, and then the calculation unit 122 calculates the daily Number of meals and total meal time.

The classification module 13 is a Bayesian classifier, and the classification module 13 further includes a feature sub-module 130 and a behavior recognition unit 133, and the feature sub-module 130 further includes a continuation unit 131 and a meal separation unit 132, wherein The feature sub-module 130 utilizes the continuation unit 131 to calculate the number of consecutive identical state features of each label data 30, and uses the consecutive identical times as the continuation feature of each label data 30, while the meal interval unit 132 further calculates the consistent state feature of each label data 30. The meal interval difference of each label material 30 of the meal precondition is determined, and according to the length of the meal interval difference, the meal interval characteristic of each label material 30 is judged as long-term or short-term. Finally, the behavior identification unit 133 uses the state feature, the round feature, the continuation state feature and the meal interval feature as variables to determine the subsequent eating behavior of the livestock 4 .

Wherein, the livestock diet monitoring system of the present invention further includes a display device 7, and the processing device 1 transmits the processing result to the display device 7, so that the animal husbandry can check the daily diet frequency and the daily total diet time of the livestock 4 through the display device 7. Eating behavior is judged and used to assess the abnormal diet of livestock 4.

Please refer to FIG. 3 , which is a flow chart of the livestock diet monitoring method of the present invention. The process of the livestock diet monitoring method of the present invention includes:

(S301) Use the detection device 3 to collect the label data 30 of the label device 2 close to the feed container 5;

(S302) Utilize the filtering unit 111 to delete the tag data 30 that is repeatedly detected and the number of times detected is too low;

(S303) Using the sorting unit 112, import the tag data 30 with the same number into the same data set 101;

(S304) Using the state unit 113, each tag data 30 is formed into an event group, and according to the difference in signal strength and the number of data records, it is determined that the state characteristic of each tag data 30 is eating or not eating;

(S305) Using the round unit 121, according to the round time difference between each label data 30 and the previous label data 30, it is judged that the round feature of each label data 30 is judged to be in-round or out-of-round;

(S306) Utilize the continuation unit 131 to calculate the number of times that the state features of each tag data 30 are consecutively the same, and take this number of times as the continuation feature;

(S307) Using the meal interval unit 132, calculate the meal interval time difference of each label material 30 that meets the meal interval prerequisite, and according to the length of the meal interval time difference, determine whether the meal interval feature of each label material 30 is long-term or short-term;

(S308) Using the calculation unit 122, evaluate the daily total number of meals and the total eating time of the livestock 4 according to the state feature and the round feature;

(S309) Using the behavior identification unit 133, using the state feature, the round feature, the continuation feature and the meal interval feature as variables, substitute the probability function (P) to obtain a similarity rate, and calculate the eating behavior of the livestock 4 according to the similarity rate.

As shown in FIG. 3, in step (S301), since the livestock diet monitoring method of the present invention is to calculate the diet status of livestock 4, the present invention utilizes the principle that livestock 4 must be close to the feed container 5 when eating and drinking. The label device 2 detected by the detection device 3 on the container 5 is assumed to be the livestock 4 is eating, and only the label data 30 of the label device 2 close to the feed container 5 is collected, so as to improve the computing efficiency of the processing device 1 , wherein the tag data 30 includes the number, signal strength and time stamp of the tag device 2 .

As shown in FIG. 3, the step (S302) filters the tag data. In order to prevent label collision and unstable signal strength, the present invention compares the codes of each label data 30 before uploading the label data 30 to the storage module 10, removes the label data 30 with duplicate codes, and sets the label at the same time. The rules for uploading the tag data 30 to the storage module 10 only after the number of detections of the device 2 reaches the collision threshold (IT).

Please refer to FIG. 4 , which is a filtering flow chart of the livestock diet monitoring method of the present invention. The steps of the present invention further comprise:

(S401) Read label data 30;

(S402) determine whether the label data 30 is read for the first time, if it is the first time to read, then execute step (S403), if it is not the first time to access then execute step (S404);

(S403) Store the label data 30 in the temporary storage unit 102;

(S404) Compare the code of the label data 30 and the label data 30 in the temporary storage unit 102, if they are the same, return to step (S401), otherwise continue to step (S405);

(S405) Store the label data 30 in the temporary storage unit 102;

(S406) Calculate the number of times that the tag device 2 is continuously detected, if the number of times exceeds the collision threshold (IT), continue to step (S407), otherwise return to step (S401);

( S407 ) Upload the label data 30 of the temporary storage unit 102 to the storage module 10 .

、

、

「cccc」、

、

、

、

、

、

。在過濾單元111比對時，在理想狀況下，僅需檢查

的前端 編碼，即可知道三串編碼是否是在儲存模組10中的標籤資料30，而不須比對整串。藉由樹狀結構過濾法，得以解決每次比對皆需將完整編碼比較過後，才得以過濾標籤資料編碼，而造成處理效率不佳之問題。 Please refer to FIG. 5A and FIG. 5B , which are schematic diagrams of a tree-like filtering method and an embodiment thereof of the livestock diet monitoring method of the present invention. The livestock diet monitoring method of the present invention improves the efficiency of the processor, and further utilizes the tree structure filtering method. Before starting to compare the label data 30, the code of the label device 2 is divided into three nodes, and then the code of the label device 2 is divided into three nodes. The node searches forward for alignment. Please refer to Table 1. In the embodiment of the present invention, at 12:00, 12:10, and 12:20, the detection device 3 detects three groups of 12-byte codes A ₁ , A ₂ , and A respectively. _3. Each group number is divided into three sections, and its format is (B ¹ , B ² , B ³ ), and B ⁱ is the data of the i-th section. In this embodiment, A ₁ = "aaaabbbbcccc", A ₂ = "aaaabbbbdddd", A ₃ = "aaaaeeeeffff", and the filtering unit 1111 uses the tree filtering method to parse into

,

,

"cccc",

,

,

,

,

,

. When the filter unit 111 compares, under ideal conditions, it is only necessary to check

The front-end code of the three strings can be known whether the three strings of codes are the label data 30 in the storage module 10 without comparing the entire strings. The tree-like filtering method can solve the problem of poor processing efficiency that the tag data code can be filtered only after the complete code needs to be compared for each comparison.

Please refer to FIG. 6 , which is a schematic diagram of an embodiment of label data classification according to the present invention. Before the present invention starts to judge the eating behavior of the livestock 4, the classifying unit 112 of the present invention firstly imports the label data 30 in the storage module 10 according to the serial number using the classifying method as described in step (S303). In different data sets 101, and arranged according to the number of times. In the embodiment of the present invention, the classifying unit 112 reclassifies the label data 30 numbered 1, 2, and 3 into data set A, data set B, and data set C, and the label data 30 in each data set 101 All are arranged in chronological order to facilitate the judgment of subsequent state characteristics.

Please refer to FIG. 7 for the processing mode of judging the state feature ( S304 ) in FIG. 3 . Fig. 7 is a flow chart of the state judgment of the livestock diet monitoring method of the present invention. In the step, the status unit 113 determines the status feature of the tag data 30 in a data set 101, and the steps include the following:

(S701) Check whether the signal strength is greater than the diet threshold (Diet Threshold, DT), if it is greater than the diet threshold (DT), execute step (S702), if it is less than, determine that its state feature is non-diet;

(S702) In the data set 101, the signal strength is greater than the dietary threshold (DT), and the label data 30 whose time sequence is continuous form an event group;

(S703) Read the label data 30; (S704) Check whether the signal strength of the label data 30 is greater than the dietary threshold (DT), if it is greater than the dietary threshold (DT), execute step (S705); (S705) Check the label data 30 Whether it is the first data in the event group to which it belongs, if it is the first data, the signal strength is temporarily stored, and the number of data in the event group is +1, otherwise, go to step (S706); (S706) Calculate the tag data 30 and the signal strength of the first tag data 30 are different; (S707) If the signal strength difference is greater than the maximum signal strength difference (ΔS ₀ ), execute step (S708 ), otherwise the number of data records of the event group to which the tag data 30 belongs is set. +1, and go back to step (S703); (S708) Check the number of data records in the event group to which the tag data 30 belongs so far, if the time interval between the data is less than the minimum duration (T ₀ ), the state of the tag data 30 will be changed. The feature is judged as diet, otherwise the state feature is judged as non-diet.

Please refer to Table 2, which is a judgment result table of an embodiment of the status feature of the present invention. In the embodiment of the present invention, the dietary threshold (DT) is 20, the maximum signal strength difference (ΔS ₀ ) is 2, and the minimum duration (T ₀ ) is 20. Among them, the label data 30 with the time sequence 3 and 4, because their signal strength does not reach the dietary threshold (DT), the state feature is judged as non-diet, and the label data 30 with the time sequence 5, 6, 7, because the time The sequence is a continuous relationship, so it is formed into an event group B, and the tag data 30 with the time sequence of 12, 13, and 14 are also composed of an event group C. Among them, the signals of the tag data 30 in the time sequence of 12, 13, and 14 The intensity difference does not exceed the maximum signal intensity difference (ΔS ₀ ), so it is judged as non-diet, and only the label data 30 in the time sequence 5, 6, and 7 pass the maximum signal intensity difference (ΔS ₀ ) and are smaller than the minimum signal intensity difference. The standard of duration (T ₀ ) can be judged as diet.

Further, the magnitude of the signal strength represents the distance between the livestock 4 and the feed container 5 . Therefore, if the signal strength is too small, it is judged that the distance from the feed container 5 is too far, and it belongs to non-diet; and the magnitude of the signal strength difference represents the movement status of the livestock 4. If it is too small, it means that livestock 4 may be resting instead of eating, and it belongs to non-diet; among them, the number of data records represents the length of time for livestock 4 to carry out an event. Since livestock 4 does not eat for too long, if an event continues If it is too long, the event may not be eating behavior, so it is judged as non-eating.

As shown in FIG. 3 , in step ( S305 ), the round unit 121 subtracts the time of the label data 30 whose state feature is diet in the same data set 101 from the time of the previous label data 30 whose state feature is diet to obtain a label The round time difference of the data 30 is determined, and then the round feature of the label data 30 whose round time difference is greater than the interval round (BCI) is judged to be out of round, otherwise it is in round. In this embodiment, the round unit 121 can be represented by the following virtual code (2):

T _i : the time when the i-th state feature is the label data 30 of diet;

BCI : Preset Interval Round (BCI);

D _i =0: the round feature of the i+1th tag data 30 is out of round;

D _i =1: The round feature of the i+1 th tag data 30 is within the round.

Please refer to Table 3, which is a judgment result table of one embodiment of the round feature of the present invention. In the embodiment of the present invention, the preset interval round (BCI) is set to 41 seconds, one round represents one meal of the livestock 4, and the round characteristic is that the label data 30 in the round represent the label data 30 and the previous label data 30 The same meal in between.

In the embodiment of the present invention, the interval round (BCI) can be derived by the interval difference method, wherein the formula of the interval difference method is as follows: formula (3):

當每日家畜4飲食總餐數的差額為零時，則表示區間回合(BCI)已將相近的標籤資料30歸為同一回合。換句話說，本發明利用設定不同的回合時間差，而找到可以使家畜4每日所飲食的餐數皆維持不變的區間回合(BCI)。

When the difference between the total meals of the daily livestock 4 is zero, it means that the interval round (BCI) has classified the similar label data 30 into the same round. In other words, the present invention finds the interval cycle (BCI) which can keep the number of meals eaten by the livestock 4 unchanged by setting different round time differences.

In the step (S306) of FIG. 3, the continuation unit 131 first determines whether the state features of the label data 30 and the previous label data 30 are the same, and if the state features are the same, the calculated state features are consecutively the same number of times, The continuum feature is this number of times. If the state feature is different, the continuation feature will be set to the initial value of 1. In this embodiment, the judgment method of the continuous state unit 131 can be represented by the following virtual code (4):

if( S _i == S _{i +1} ){ L=L+1; }else if( S _i != S _{i +1} ){ L=1; } L _i = L ; (4)

S _i : the state feature of the i-th tag data 30;

L: Continuum feature.

Please refer to FIG. 8 for the processing mode of judging the interval meal feature ( S307 ) in FIG. 3 , which is a flowchart of the present invention for judging the interval meal feature. In the step (S307) of the present invention, the meal interval unit 132 calculates, when the label data 30 is detected, the meal interval time that the livestock 4 has elapsed, and assigns the label data 30 a corresponding meal interval based on the length of the meal interval time. Meal interval characteristics. The steps of the present invention further comprise:

(S801) Read label data 30;

(S802) Determine whether the state feature of the label data 30 is diet, whether the round indicator is within the round, whether the continuation state feature is greater than the continuous threshold (CT) and other preconditions for meal separation, and if the above preconditions for meal separation cannot be fully met, the The detected time of the label data 30 is set as the initial time, and the meal-interval feature is set as short-term, if the above-mentioned preconditions for meal-interval are met, execute step (S803);

(S803) Calculate the time difference between the time when the tag data 30 is detected and the initial time, and take the time difference as the time difference between meals, and determine whether the time difference between meals is greater than the time between meals (TBM), if If it is greater than the time between meals (TBM), the interval between meals is set as long-term, otherwise it is set as short-term.

3, in the step (S305) of the present invention, after the round unit 121 extracts the round features of each tag data 30, in the step (S307) of the present invention, the calculation unit 122 uses the round features to infer each round feature of the livestock 4. The number of rounds (meals) per day, or what is known as the number of meals per day, and based on the time difference between rounds of each meal, the total daily eating duration of the livestock 4 is calculated.

Please refer to Table 3 again, in this example, the daily number of meals of livestock 4 is 2 meals, and the total daily duration of meals is "the sum of the round time difference of the first meal and the round time difference of the second meal", The sum of the round time difference of the first meal is 20+20+40+20+20+20=140 (minutes), the sum of the round time difference of the second meal is 20+20=40 (minutes), and then the first meal The sum of the bout time differences from the second meal gives the total daily diet duration of Animal 4 to be 180 minutes.

In the step (S308) of the present invention, the behavior identification unit 133 of the present invention uses a probability function (P) to calculate the similarity rate of the eating behavior of the livestock 4, and if the similarity rate is greater than 1, it is determined that the livestock 4 will eat. If the similarity rate is less than 1, it is judged that the livestock 4 will not eat, and the probability function (P) is the following formula (1):

LH: similarity rate;

L _i : the continuation feature of the i -th tag data 30;

D _i : the round feature of the i -th tag data 30;

f _i : the meal interval feature of the i -th label data 30;

S _i : the state feature of the i -th tag data 30;

S _{i +1} =1: the state feature of the i +1 th tag data 30 is diet;

S _{i +1} =0: The state characteristic of the i +1 th tag data 30 is non-diet.

More specifically, P ( S _{i +1} =1| L _i , D _i , f _i , S _i ) is when the i -th label data 30 continuation feature is L _i , the round feature is D _i , and the meal interval feature is When is f _i and the state feature is S _i , the state feature of the i +1th tag data 30 is the probability of eating, and P ( S _{i +1} =0| Li ,D _i ,fi , S _{i )} is the probability that the state feature of the i +1 th tag data 30 is non-diet under the same conditions. If the similarity ratio is greater than 1, it means that the probability that the state feature of the i +1 th tag data 30 is diet is greater than the probability of non-diet.

In the livestock diet monitoring system and method of the present invention, the processing device 1 , the labeling device 2 and the detection device 3 can accurately record the daily diet frequency and total diet duration of the livestock 4 , and further according to the behavior pattern and observation of the livestock 4 Record, design the criteria for the state feature, round feature, continuation feature and meal interval feature, and use the above features as variables, use the probability model constructed by Bayes' theorem to calculate the similarity rate of the follow-up behavior of livestock 4 Find the follow-up diet The possibility of behavior, so as to achieve the purpose of judging livestock 4 diet monitoring.

The above detailed descriptions are specific descriptions of feasible embodiments of the present invention, but the embodiments are not intended to limit the patent scope of the present invention. Any equivalent implementation or modification that does not depart from the technical spirit of the present invention shall be included in this case. within the scope of the patent.

S301-S309: Steps

Claims (10)

A livestock diet monitoring method is applied to a breeding space, wherein the breeding space comprises a feed container, a detection device and a target provided with a label device, and the detection device is arranged around the feed container, wherein the livestock diet Monitoring methods include: Use the detection device to detect the label device within its detection range to obtain a label data, wherein each label data includes the serial number, signal strength, and detected time of each label device; Using a categorization unit, import each of the tag data with the same number into a data set in chronological order; Using a state unit, each of the tag data is formed into an event group, and the signal intensity difference and the number of data records of each event group are calculated, and each tag data is assigned a state feature, wherein if the signal intensity difference of the tag data is different If the maximum signal strength difference (ΔS ₀ ) is met, and the interval between the two pieces of data is less than the minimum duration (T ₀ ), the state feature of the label data is set as diet, otherwise, it is non-diet; Using a round unit, calculate the round time difference of each label data whose state feature is diet, and assign a round feature, wherein the round feature of each label data whose round time is less than the interval round (BCI) is within round, otherwise out of the round; Using a computing unit, the total eating time of the target is estimated according to the time difference between the rounds, and the total eating times of the target is estimated according to the characteristics of the round. The livestock diet monitoring method as claimed in claim 1, wherein the event group is composed of each of the tag data in each of the data sets whose time sequence is consecutive. The livestock diet monitoring method as claimed in claim 1, wherein the signal strength difference is the signal strength difference between the tag data and the first tag data in the event group to which it belongs, wherein the number of data records is the total number of the tag data in the event group up to the time point when the tag data is detected in each event group. The livestock diet monitoring method as claimed in claim 1, wherein the round time difference is the time between each label data whose state feature is diet and the previous label data whose state feature is diet in each of the data sets difference. The method for monitoring the diet of livestock as described in item 1 of the scope of the patent application further comprises using a filtering unit to compare the data of each of the tags, to remove the repeated data of each of the tags, and to calculate the number of times that each of the tags is continuously detected, and Remove each tag data whose number of consecutive detected times does not meet the collision threshold (IT). The livestock diet monitoring method according to claim 1, further comprising using a feature sub-module to perform a feature extraction method step, the feature extraction method comprising: using a continuation unit to assign a continuation feature to each of the tag data, wherein If each of the label data is the same as the state feature of the previous label data, the number of times that the state feature is consecutively the same is calculated, and the continuation state feature is the number of times, where if the label data is the same as the previous label data If the state feature of the label data is different, the continuation state feature is 1; each of the label data is assigned a meal-separation feature by using a meal-separation unit. The meal interval difference of the label data, and the meal interval feature of each label data whose meal interval time difference is greater than the time between meals (Time Between Meals, TBM) is set as long-term, otherwise it is short-term, and its If each of the label data does not meet the pre-conditions of the meal interval, set the meal interval feature of each label data as short-term, and take the time when each label data is detected as an initial time; wherein, the interval meal Prerequisites include that the state feature of the tag data is diet, the round feature is in-round, and the continuation state feature is greater than a continuous threshold (CT); wherein, the time difference between meals is the time each of the tag data is detected. The time difference between its closest initial time. The livestock diet monitoring method according to claim 6, further comprising using a behavior identification unit to perform a judgment method for judging the eating behavior of the target, the judgment method comprising: using a probability function (P) to determine the target's eating behavior. The state feature, the round feature, the continuation feature, and the meal interval feature are variables, and a similarity ratio is calculated. If the similarity ratio is greater than 1, it is determined that the target will eat and drink, and if the similarity ratio is less than 1, it is determined that The target will not eat and drink, where the probability function (P) is the formula (1) listed below:

Among them, LikelyHood represents the similarity ratio, Li represents the continuation feature of the ith tag data of the target, D _i represents the round feature of the _ith tag data of the target, and f _i represents the ith stroke of the target The meal interval feature of the tag data, S _i represents the state feature of the i -th tag data of the target, S i ₊₁ =1 represents the i-th +1-th item of the target The state feature of the tag data is diet, S _{i +1} =0 indicates that the i +1 th tag data of the target has a state feature of non-diet. A livestock diet monitoring system is applied to a breeding space, wherein the breeding space includes a feed container and a target, and the livestock diet monitoring system includes: a label device, the label device is arranged on the target; a detection device, The detection device is arranged around the feed container, and the detection device is used for the detection device to detect the label device within its detection range to obtain label data of the label device, wherein each label data includes the label data of the label device. serial number, signal strength, and detected time; a display device, a processing device, connected with the detection device and the display device, wherein the processing device further includes: a classification unit for the serial number The same tag data are imported into a data set in chronological order; a state unit is used to form each tag data into an event group, and calculate the signal strength difference and the number of data records of each event group, and assign the data to each event group. A state feature of the tag data, wherein if the signal strength difference of one of the event groups to which the tag data belongs is greater than the maximum signal strength difference (ΔS ₀ ), and the interval between two pieces of data of the data is less than the minimum duration (T ₀ ) , then the label data is judged as diet, otherwise it is non-diet; a round unit is used to calculate the round time difference of each label data whose state feature is diet, and assign a round feature, where the round time difference is smaller than the interval round (BCI) The characteristic of the round of each of the label data is inside the round, otherwise it is outside the round; a calculation unit, according to the time difference of the round, and evaluate the total eating time of the target, and according to the characteristics of the round, and evaluate the total eating time of the target number of meals. A livestock diet monitoring system as claimed in claim 8, further comprising a feature sub-module for performing the feature extraction method as claimed in claim 6, wherein the feature sub-module further comprises: a continuous state unit for using In assigning each of the label data a continuation feature, wherein if each of the label data is the same as the state feature of the label data preceding it, the number of times the state feature is consecutively identical is calculated, and the continuation feature is the The number of times, wherein if the label data is different from the state feature of the previous label data, the continuation state feature is set to the initial value of 1; a meal interval unit is used to give each label data a meal interval feature , if each of the label data complies with the pre-condition of a meal interval, then calculate the meal interval difference of each of the label data, and calculate the interval of each label data whose meal interval time difference is greater than the time between meals (Time Between Meals, TBM). The meal feature is set to long-term, otherwise it is short-term. If the label data does not meet the pre-conditions for meal separation, the meal separation feature of each label data is set to short-term, and the time when each label data is detected is used. is an initial time; wherein, the precondition of the meal interval includes that the state feature of the label data is diet, the round feature is in-round, and the continuation state feature is greater than the continuous threshold (CT); Wherein, the time difference between meals is the time difference between the time when each of the tag data is detected and the most recent initial time. A livestock diet monitoring system as claimed in claim 8, further comprising a behavior recognition unit for executing the judgment method as claimed in claim 7.

Images