Deep learning-based beef cattle limp behavior detection system and method

Abstract

The invention discloses a beef cattle limp behavior detection system and method based on deep learning, the system comprises a data acquisition module, a data transmission module, a data analysis module and a mobile receiving module, and the method comprises the following steps: acquiring walking data and positioning data of beef cattle in real time through the data acquisition module; the positioning data and the walking data are sent to a data analysis module through a data transmission module; the positioning data are analyzed through the data analysis module, positioning coordinates are generated, a limp behavior detection model is established in combination with the walking data, the walking behavior of the beef cattle is recognized in real time through the limp behavior detection model, and limp anomaly detection is conducted according to the walking data; and the analyzed positioning coordinates and the limping anomaly detection result are sent to a mobile receiving module, so that accurate positioning and abnormal behavior early warning of beef cattle are realized. According to the method, the manual workload is reduced, the accuracy of judging the limp behavior of the beef cattle is improved, the cost of a farm can be reduced to a certain extent, and the animal welfare is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of beef cattle breeding, and in particular, relates to a beef cattle lameness behavior detection system and method based on deep learning. Background Art

[0002] In the development of modern agriculture and animal husbandry, the application of digital and intelligent technologies is becoming more and more popular. By using advanced sensor technology and wireless communication technology, farmers and ranchers can monitor and manage the health and behavior of livestock in real time, thereby improving production efficiency and animal welfare. The lameness behavior of beef cattle directly affects the health of beef cattle, thereby affecting the productivity and economic benefits of the ranch. Traditional manual monitoring methods are not only time-consuming and labor-intensive, but also have poor accuracy and timeliness. Automated monitoring through technical means can promptly detect abnormal conditions such as disease, injury or malnutrition, so that corresponding measures can be taken to reduce economic losses and improve production efficiency.

[0003] Therefore, how to monitor the lameness behavior of beef cattle in real time through digital and intelligent technology and further improve the accuracy and timeliness of monitoring is a problem that needs to be urgently solved in the existing technology. Summary of the invention

[0004] The purpose of the present invention is to provide a beef cattle lameness behavior detection system and method based on deep learning, so as to overcome the defects of the prior art that the manual monitoring method for beef cattle lameness behavior monitoring is not only time-consuming and labor-intensive, but also has poor accuracy and timeliness.

[0005] The present invention is achieved by adopting the following technical solutions: A beef cattle lameness behavior detection system based on deep learning, comprising: The data acquisition module is used to obtain the walking data and positioning data of beef cattle in real time, and store the walking data after filtering; A data transmission module, used for wirelessly sending walking data and positioning data to a data analysis module; The data analysis module is used to analyze the positioning data, generate positioning coordinates, and establish a lameness behavior detection model in combination with the walking data. The BOHB optimization algorithm is used to find the optimal hyperparameter combination. The lameness behavior detection model is used to identify the walking behavior of beef cattle in real time and perform lameness abnormal behavior detection. The information receiving module is used to receive the walking information and positioning information analyzed by the data analyzing module.

[0006] Furthermore, the data acquisition module includes a sensor device and a Bluetooth beacon, the sensor device includes a shell and a nine-axis posture sensor, a Bluetooth module, a data storage unit, a main control chip and a power supply unit integrated in the shell, the Bluetooth beacon is used to broadcast signals in real time, the nine-axis posture sensor is used to collect walking data of beef cattle in real time, the data storage unit is used to store the walking data collected by the nine-axis posture sensor, and the Bluetooth module is used to scan and receive positioning data of the Bluetooth beacon; the main control chip is used for calling the nine-axis posture sensor, the Bluetooth module, and the data storage unit and processing data; the power supply unit is used to power the nine-axis posture sensor, the Bluetooth module, the data storage unit, and the main control chip.

[0007] Furthermore, the data storage unit includes an SD memory card.

[0008] Furthermore, the data transmission module includes a Lora unit and a Lora base station. The Lora unit is electrically connected to the main control chip and is used to send the positioning data received by the Bluetooth module and the walking data monitored by the nine-axis attitude sensor to the Lora base station via wireless signals. The Lora base station is used to transmit the data sent by the Lora unit to the data analysis module via wireless signals.

[0009] Furthermore, the lameness behavior detection model uses the BOHB optimization algorithm to find the optimal hyperparameter combination step including: First, the data is organized by normalization preprocessing method to convert the sample data into mean , standard deviation , a normal distribution, each feature in the sample undergoes the following transformation: , The length of each batch of data input to the network is divided into three levels. Each batch of data corresponds to a sliding window. The first layer is Seq_len, which is used as the input of the Encoder layer to encode and extract features from past time steps. Position Encoding (PE) combines feature vector and position information through the following formula: , in, POS It means that the data can be located at a certain time step. d is the dimension of the position encoding, 2 i and 2 i +1 for 2 i ≤ d , 2 i +1≤ d The even and odd dimensions of Extract feature vectors through position encoding (PE) : , in, n is the number of samples.

[0010] The second layer is Label_len, which is the input of the decoder layer. Value encoding (VE) is used to capture more information feature values. The changes after encoding are as follows: , in, n is the number of samples, d is the dimension, which is consistent with the dimension of the eigenvector after PE.

[0011] The third layer is Pred_len, which is also the input of the decoder layer. TE is used to record the feature arrangement information of the entire sequence at each time step to enhance the time correlation between data at different times in the sliding window. The TE conversion process is as follows: , The three eigenvectors The sum is: , as input data for the model.

[0012] Furthermore, the lameness behavior detection model real-time identification of the walking behavior algorithm of beef cattle includes: The input data is divided into X en and X de ,After passing the embedding block, the features are fed into the encoder and decoder blocks, the output of the encoder is organized through the attention mechanism of the decoder block, which is then passed to the LSTM block and the fully connected layer, and the results are output in parallel.

[0013] The present invention also provides a method for detecting lameness behavior of beef cattle based on deep learning, based on the above-mentioned lameness behavior detection system of beef cattle based on deep learning, comprising the following steps: The walking data and positioning data of the beef cattle are acquired in real time through the data acquisition module, and the walking data is stored in the data storage module; The positioning data and the walking data in the data storage module are sent to the data analysis module through the data transmission module; The data analysis module analyzes the positioning data, generates positioning coordinates, and establishes a lameness behavior detection model in combination with the walking data. The lameness behavior detection model is used to identify the walking behavior of beef cattle in real time, and lameness abnormality detection is performed based on the walking data. The parsed positioning coordinates and lameness abnormality detection results are sent to the mobile receiving module to achieve accurate positioning of beef cattle and abnormal behavior warning.

[0014] Compared with the prior art, the present invention has the following beneficial effects: (1) The present invention collects reliable walking data and positioning data of beef cattle by installing data collection modules on beef cattle and cowsheds, thereby obtaining more accurate physiological activity data of beef cattle. At the same time, a lameness behavior detection model is established in combination with the walking data. The walking behavior of beef cattle is identified in real time through the lameness behavior detection model, and lameness abnormal behavior detection is performed. The lameness abnormality detection results and positioning information are sent to a mobile receiving module via the cloud, thereby realizing abnormal behavior warning of beef cattle, reducing manual workload, and improving the accuracy of judging lameness behavior of beef cattle. To a certain extent, the cost of breeding farms can be reduced and animal welfare can be improved. (2) The present invention uses Bluetooth and Lora fusion positioning technology, combining the advantages of both. Bluetooth has the advantages of low power consumption, short-range communication and high data transmission rate, and can perform high-precision positioning and data transmission in a local range. Lora technology has the advantages of long-distance transmission and low power consumption. The combination of the two can provide reliable positioning and long-distance data transmission. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 The present invention is a schematic diagram of the process of lameness behavior detection of a beef cattle lameness behavior detection system based on deep learning.

[0016] Figure 2 This is a flow chart of the positioning principle of a beef cattle lameness behavior detection system based on deep learning in the present invention.

[0017] Figure 3 This is a structural block diagram of a data acquisition module of a beef cattle lameness behavior detection system based on deep learning in the present invention.

[0018] Figure 4 The present invention is a schematic diagram of the algorithm flow of a detection model of a beef cattle lameness behavior detection system based on deep learning.

[0019] In the figure: 1. Sensor device; 2. Lora base station; 3. Back-end server; 4. Mobile terminal; 5. Cloud server; 6. Bluetooth beacon; 7. Power supply unit; 8. Bluetooth module; 9. Main control chip; 10. Nine-axis attitude sensor; 11. Data storage unit; 12. Lora unit. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the examples of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] The beef cattle lameness behavior detection system based on deep learning of the present invention comprises a data acquisition module, a data transmission module, a data analysis module and a mobile receiving module, wherein: The data acquisition module is used to obtain the walking data and positioning data of beef cattle in real time, and store the walking data after filtering.

[0022] like Figures 1 to 3 As shown, the data acquisition module includes a sensor device 1 and a Bluetooth beacon 6. The sensor device 1 includes a shell and a nine-axis posture sensor 10, a Bluetooth module 8, a data storage unit 11, a main control chip 9 and a power supply unit 7 integrated in the shell. The Bluetooth beacon 6 is used to broadcast signals in real time, the nine-axis posture sensor 10 is used to collect walking data of beef cattle in real time, the data storage unit 11 is used to store the walking data collected by the nine-axis posture sensor 10, and the Bluetooth module 8 is used to scan and receive the positioning data of the Bluetooth beacon; the main control chip 9 is used for calling the nine-axis posture sensor 10, the Bluetooth module 8, and the data storage unit 11 and processing the data; the power supply unit 7 is used for powering the nine-axis posture sensor 10, the Bluetooth module 8, the data storage unit 11, and the main control chip 9.

[0023] Among them, the data storage unit 11 includes an SD memory card, which is used as a data storage device to save data and facilitate data transmission and reception and later reading; the sensor device 1 uses STM32u575 as the main control chip 9 of the sensor, communicates and calls with the Bluetooth module 8 and the nine-axis attitude sensor 10 through the SPI communication protocol, stores the data collected by the nine-axis attitude sensor 10 to the SD card, and is powered by the power supply unit 7.

[0024] When the nine-axis posture sensor 10 starts working, it generates data in real time, sets a time interval, and stores the data generated by the nine-axis posture sensor 10 into the SD card at regular intervals. It sets a dynamic threshold, and the nine-axis posture sensor 10 is activated only when the accelerometer in the nine-axis posture sensor 10 changes more than a certain range. When it is turned off, it is determined whether to stand or lie down based on the generated walking data.

[0025] The data transmission module is used to send the positioning data and the walking data stored in the data storage module 11 to the data analysis module.

[0026] The data transmission module includes a Lora unit 12 and a Lora base station 2. The Lora unit is electrically connected to the main control chip 9 and integrated in the sensor device 1. It is used to send the positioning data received by the Bluetooth module 8 and the walking data monitored by the nine-axis posture sensor 10 to the Lora base station 2 via wireless signals. The Lora base station 2 is used to transmit the data sent by the Lora unit 12 to the data analysis module via wireless signals.

[0027] During data transmission, the walking data is sent to the Lora base station 2 by calling the Lora unit 12 module through the main control chip 9 via wireless signals, and the Lora base station 2 sends it to the back-end server 3 of the data analysis module via wireless transmission.

[0028] The data analysis module is used to analyze the positioning data, generate positioning coordinates, and establish a lameness behavior detection model in combination with the walking data. The BOHB optimization algorithm is used to find the optimal hyperparameter combination. The lameness behavior detection model is used to identify the walking behavior of beef cattle in real time and perform lameness abnormal behavior detection.

[0029] The method for finding the optimal hyperparameter combination using the BOHB optimization algorithm for the lameness behavior detection model includes the following steps: First, the data is organized by normalization preprocessing method to convert the sample data into mean , standard deviation , a normal distribution, each feature in the sample undergoes the following transformation: , The length of each batch of data input to the network is divided into three levels. Each batch of data corresponds to a sliding window. The first layer is Seq_len, which is used as the input of the Encoder layer to encode and extract features from past time steps. Position Encoding (PE) combines feature vector and position information through the following formula: , in, POS It means that the data can be located at a certain time step. d is the dimension of the position encoding, 2 i and 2 i +1 for 2 i ≤ d , 2 i +1≤ d The even and odd dimensions of Extract feature vectors through position encoding (PE) : , in, n is the number of samples.

[0030] The second layer is Label_len, which is the input of the decoder layer. Value encoding (VE) is used to capture more information feature values. The changes after encoding are as follows: , in, n is the number of samples, dis the dimension, which is consistent with the dimension of the eigenvector after PE.

[0031] The third layer is Pred_len, which is also the input of the decoder layer. TE is used to record the feature arrangement information of the entire sequence at each time step to enhance the time correlation between data at different times in the sliding window. The TE conversion process is as follows: , The three eigenvectors The sum is: , as input data for the model.

[0032] It should be noted that the BOHB algorithm is a combination of the Bayesian optimization algorithm and the hyperband algorithm. Under the hyperband architecture, the Bayesian optimization algorithm is used to replace the random search algorithm of the successive halving algorithm in each step.

[0033] Among them, Figure 4 As shown in the figure, the lameness behavior detection model real-time identification of beef cattle walking behavior algorithm includes: the input data is divided into X en and X de ,After passing the embedding block, the features are fed into the encoder and decoder blocks, the output of the encoder is organized through the attention mechanism of the decoder block, which is then passed to the LSTM block and the fully connected layer, and the results are output in parallel.

[0034] The mobile receiving module is used to receive the detection information and positioning coordinates analyzed by the data analyzing module.

[0035] The mobile receiving module includes a mobile terminal 4 and a cloud server 5. The detected walking information and positioning coordinate information are sent to the mobile terminal 4 through the cloud server 5. The mobile terminal 4 adopts a smart mobile device such as a smart phone.

[0036] Principle of lameness behavior detection of the present invention: Figure 1 The schematic diagram of the workflow of lameness behavior detection of the present invention is as follows: first, sensor devices 1 are worn on multiple beef cattle, and the sensor devices 1 collect walking data of the beef cattle in real time and perform filtering processing. The sensor device 1 performs labeling processing, and packages the walking data and transmits it to the Lora base station 2 by wireless transmission. The Lora base station 2 acts as a middleware and acts as a bridge, and transmits the data to the back-end server 3 by wireless transmission. The BOHB-Informer-LSTM model of the back-end server 3 performs lameness behavior detection, and records the number and occurrence time of the behavior as the standard for the model to detect abnormalities of beef cattle. The walking information of the beef cattle is sent to the mobile terminal 4 in the form of a cloud server 5, which is convenient for the staff to make decisions and process.

[0037] The positioning principle of the present invention: Figure 2 The flowchart of the positioning principle of the present invention is shown as follows. A Bluetooth beacon 6 is installed at the fence, and a Bluetooth AOD positioning method is adopted. The transmitting end uses the Bluetooth beacon 6, and the receiving end is a Bluetooth module 8 embedded in the sensor device 1. The Bluetooth beacon 6 installed on the fence of the cowshed broadcasts the signal in real time. The Bluetooth module 8 scans the signal broadcast by the Bluetooth beacon 6, and sends it to the Lora base station 2 through the wireless signal of the Lora module. The Lora base station 2 sends it to the back-end server 3 of the data analysis module through wireless transmission. The positioning data is analyzed on the back-end server 3, and the positioning coordinates after the analysis are sent to the mobile receiving module 4 through the cloud, so as to realize the accurate positioning of the sensor and generate a map. When a beef cattle with abnormal behavior is found, the mobile terminal 4 of the mobile receiving module receives an early warning, and determines the cowshed to which it belongs according to the number of the beef cattle. The veterinarian searches for abnormal signals according to the map of the mobile terminal 4 and handles them.

[0038] The present invention also provides a method for detecting lameness behavior of beef cattle based on deep learning, comprising the following steps: S1. Obtain the walking data and positioning data of beef cattle in real time through the data acquisition module, and store the walking data in the data storage module; S2. The positioning data and walking data in the data storage module are sent to the data analysis module through the data transmission module; S3. Analyze the positioning data through the data analysis module, generate positioning coordinates, establish a lameness behavior detection model in combination with the walking data, identify the walking behavior of beef cattle in real time through the lameness behavior detection model, and perform lameness abnormality detection based on the walking data; S4. Send the resolved positioning coordinates and lameness abnormality detection results to the mobile receiving module to achieve accurate positioning of beef cattle and abnormal behavior warning.

Claims

1. A beef cattle lameness behavior detection system based on deep learning, characterized in that: include: The data acquisition module is used to obtain the walking data and positioning data of beef cattle in real time, and store the walking data after filtering; A data transmission module, used for wirelessly sending walking data and positioning data to a data analysis module; The data analysis module is used to analyze the positioning data, generate positioning coordinates, and establish a lameness behavior detection model in combination with the walking data. The BOHB optimization algorithm is used to find the optimal hyperparameter combination. The lameness behavior detection model is used to identify the walking behavior of beef cattle in real time and perform lameness abnormal behavior detection. The mobile receiving module is used to receive the walking information and positioning information analyzed by the data analyzing module.

2. A beef cattle lameness behavior detection system based on deep learning according to claim 1, characterized in that: The data acquisition module includes a sensor device and a Bluetooth beacon. The sensor device includes a shell and a nine-axis posture sensor, a Bluetooth module, a data storage unit, a main control chip and a power supply unit integrated in the shell. The Bluetooth beacon is used to broadcast signals in real time. The nine-axis posture sensor is used to collect walking data of beef cattle in real time. The data storage unit is used to store the walking data collected by the nine-axis posture sensor. The Bluetooth module is used to scan and receive positioning data of the Bluetooth beacon; the main control chip is used for calling the nine-axis posture sensor 10, the Bluetooth module 8, and the data storage unit 11 and processing data; the power supply unit 7 is used to supply power to the nine-axis posture sensor 10, the Bluetooth module 8, the data storage unit 11, and the main control chip 9.

3. A beef cattle lameness behavior detection system based on deep learning according to claim 2, characterized in that: The data storage unit includes an SD memory card.

4. A beef cattle lameness behavior detection system based on deep learning according to claim 2, characterized in that: The data transmission module includes a Lora unit and a Lora base station. The Lora unit is electrically connected to the main control chip and is used to send the positioning data received by the Bluetooth module and the walking data monitored by the nine-axis attitude sensor to the Lora base station via wireless signals. The Lora base station is used to transmit the data sent by the Lora unit to the data analysis module via wireless signals.

5. The beef cattle lameness behavior detection system based on deep learning according to claim 1, characterized in that: The lameness behavior detection model uses the BOHB optimization algorithm to find the optimal hyperparameter combination steps including: First, the data is organized by normalization preprocessing method to convert the sample data into mean , standard deviation , a normal distribution, each feature in the sample undergoes the following transformation: ， The length of each batch of data input to the network is divided into three levels. Each batch of data corresponds to a sliding window. The first layer is Seq_len, which is used as the input of the Encoder layer to encode and extract features from past time steps. Position Encoding (PE) combines feature vector and position information through the following formula: ， in, POS It means that the data can be located at a certain time step. d is the dimension of the position encoding, 2 i and 2 i +1 for 2 i ≤ d , 2 i +1≤ d The even and odd dimensions of Extract feature vectors through position encoding (PE) : , in, n is the number of samples, The second layer is Label_len, which is the input of the decoder layer. Value encoding (VE) is used to capture more information feature values. The changes after encoding are as follows: ， in, n is the number of samples, d is the dimension, which is consistent with the dimension of the eigenvector after PE. The third layer is Pred_len, which is also the input of the decoder layer. TE is used to record the feature arrangement information of the entire sequence at each time step to enhance the time correlation between data at different times in the sliding window. The TE conversion process is as follows: ， The three eigenvectors The sum is: ， as input data for the model.

6. The beef cattle lameness behavior detection system based on deep learning according to claim 1, characterized in that: The lameness behavior detection model real-time identification of beef cattle walking behavior algorithm includes: The input data is divided into X en and X de ,After passing the embedding block, the features are fed into the encoder and decoder blocks, the output of the encoder is organized through the attention mechanism of the decoder block, which is then passed to the LSTM block and the fully connected layer, and the results are output in parallel.

7. A method for detecting lameness behavior of beef cattle based on deep learning, characterized in that: A beef cattle lameness behavior detection system based on deep learning according to any one of claims 1 to 6, comprising the following steps: The walking data and positioning data of the beef cattle are acquired in real time through the data acquisition module, and the walking data is stored in the data storage module; The positioning data and the walking data in the data storage module are sent to the data analysis module through the data transmission module; The data analysis module analyzes the positioning data, generates positioning coordinates, and establishes a lameness behavior detection model in combination with the walking data. The lameness behavior detection model is used to identify the walking behavior of beef cattle in real time, and lameness abnormality detection is performed based on the walking data. The parsed positioning coordinates and lameness abnormality detection results are sent to the mobile receiving module to achieve accurate positioning of beef cattle and abnormal behavior warning.

Images