Shobowale, Rukayat Temitope, Baridam, Barileé Barisi,
References
iv. Cloud-based platform: The cloud-based platform functions as a data processing and analysis centre, processing and analysing sensor data using advanced analytics and machine learning techniques. v. Dashboard: Provision was made for real-time data visualisation of the animal's health state via the dashboard, allowing animal owners and veterinarians to interact and share ideas while monitoring the animal's health and making educated decisions. vi. Mobile application: A mobile application was created to allow animal owners and vets to remotely monitor the animal's health, get alarms, and access historical data. The methodology included the collection of data/parameters for anorexia (loss of appetite), abdominal pain, aggression, coughing, depression, diarrhoea, dehydration, lameness (difficulty walking), lethargy (lack of energy), fever, nausea, pain, pneumonia, rapid breathing, reduced fertility, salivation (excessive drooling), swelling in various body parts (e.g., pharyngeal, udder), tachycardia (elevated heart rate), weight loss, and weakness The suggested CNN IoT-based enhanced monitoring system for animal health consists of numerous components, including temperature sensors, weight sensors, sound sensors, activity sensors, and a microcontroller module, also known as the Arduino platform. Some of these sensors were attached to the animal's body to collect data on the animal's health parameters, while others were attached to their pens and surrounding environments to monitor activities within and outside the animal's immediate environment, allowing for real-time and rapid response to all threats and abnormalities. The microcontroller module analyses these data before sending them to a central server or cloud-based platform for visualisation and analysis. One of the key advantages of adopting IoT technology to monitor animal health is the ability to gather data remotely and continuously. This eliminates the need for manual data collection, which often times, are time- consuming, labour-intensive, and data-inconsistent or mismatched. Furthermore, CNN IoT-based systems can detect the slightest variation in an animal's health parameters that may be unnoticed by humans. The TeSLF Embedded system has several potential applications, including monitoring livestock health in agriculture, tracking animal health domestically, in zoos, or in wildlife scenarios; it also preserves and facilitates animal behaviour and physiology research. Furthermore, the system was utilised to prevent and detect the spread of animal illnesses, which could have serious economic and public health consequences. Architecture of the TeSLF Embedded system Figure 11: Architecture of the TeSLF Embedded system Figure 12: Circuit Diagram of the TeSLF Embedded system Software Requirements 1. Operating System: A lightweight operating custom OS for specialized devices system suitable for Arduino Nano BLE Sense was installed. 2. Deep Learning Framework – TensorFlow 3.2.0, Arduino IDE 2.2.1 were deployed on the Arduino Nano Ble Sense 3. Dataset: The data used were live data from the field. 5. Training: Teachable Machines from Google were deployed on the machine to train the data. PB0/ICP1/CLKO/PCINT0 14 PB1/OC1A/PCINT1 15 PB3/MOSI/OC2A/PCINT3 17 PB2/SS/OC1B/PCINT2 16 PD6/AIN0/OC0A/PCINT22 12 PD5/T1/OC0B/PCINT21 11 PD4/T0/XCK/PCINT20 6 PD3/INT1/OC2B/PCINT19 5 PD2/INT0/PCINT18 4 PD1/TXD/PCINT17 3 PD0/RXD/PCINT16 2 PB4/MISO/PCINT4 18 PB5/SCK/PCINT5 19 PB7/TOSC2/XTAL2/PCINT7 10 PB6/TOSC1/XTAL1/PCINT6 9 PC6/RESET/PCINT14 1 PC5/ADC5/SCL/PCINT13 28 PC4/ADC4/SDA/PCINT12 27 PC3/ADC3/PCINT11 26 PC2/ADC2/PCINT10 25 PC1/ADC1/PCINT9 24 PC0/ADC0/PCINT8 23 AVCC 20 AREF 21 PD7/AIN1/PCINT23 13 U2 ATMEGA328P 3 2 7 4 6 1 5 8 U5 CA3140 3 2 7 4 6 1 5 8 U6 CA3140 CS 1 VIN(+) 2 VIN(-) 3 GND 4 VCC 8 CLK 7 VREF 5 DO 6 U7 27.0 DQ 1 CLK/CONV 2 RST 3 THIGH 7 TLOW 6 TCOM 5 U8 27.0 CE 2 SCK 3 SDO 5 SDI 6 U3 SCL 6 SDA 7 AD0 8 AD1 9 A1 1 W1 2 B1 3 O1 14 O2 12 SHDN 5 VDD 4 GND 10 VSS 11 U1 GPS Module SCL 6 SDA 7 AD0 8 AD1 9 A1 1 W1 2 B1 3 O1 14 O2 12 SHDN 5 VDD 4 GND 10 VSS 11 U9 AVDD 24 GND VCC 2 SCLK 8 SDIN 9 LATCH 7 SDO 10 CLEAR 6 DVCC SEL 16 REFOUT 14 REFIN 15 SENSE 18 BOOST 20 IOUT 19 FAULT 3 RSET 13 CAP1 22 CAP2 21 U4 ESP32 VI 1 VO 3 GND 2 U10 7805 VI 1 VO 3 GND 2 U11 7805 VCC D+ D- GND J1 C1 33u D1 DIODE C2 1u C3 1u temperature and Humidity Sensor (Enviroment) (Arduino Nano) GPS Transciever R1 470k R2 11k High Precision A/D Converter USB Charging Port Mic 27.0 3 1 VOUT 2 U12 LM34 Body Termometer 6. Classification: Convolutional Neural Network was used for prediction. 7. Database: MySQL Engine was integrated with the Teachable Machine. 8. User Interface: The ThingSpeak IOT App was installed and integrated to the edge device for remote viewing. Hardware Requirements 1. Arduino Nano BLE Sense Microcontroller 2. Solar Panel 3. DS18B20 temperature sensor 4. Digital Humidity and Temperature Sensor 5. Omnidirectional Condenser Microphone 6. 3.7v 3300mA Lithium Ion Batteries 7. TD4056 Smart Lithium Battery Charger IC 8. Electronic Scale Training and Testing Modules of the TeSLF Embedded system Training Module of the TeSLF Embedded system Mathwork is a web tool that makes it fast and easy to create machine learning models for a project. These models were trained using a learning technique called transfer learning. This is a machine learning (ML) method where a model trained on one tasklearns to be implemented for other tasks, similar to the functionality of a Case-Based Reasoning system . Basically, a neural network is trained on a large dataset and retains its knowledge for future use. This approach not only requires less computing power but also requires a smaller dataset for training, having been trained earlier. The Mathwork platform currently supports three types of models, which are Pose, Sound and Image classifiers. The sound and pose classifiers are the core component of this research. The Teachable training often involves hundreds of images and data on the dataset to arrive at an accurate prediction based on the volume of the dataset and the number of hidden layers available. This training is called MobileNet. Step 1: Open the URL: https://teachablemachine.withgoogle.com/train/image on Google Chrome browser, create a new project by clicking on image. Step 2: On the Google Teachable Machines Page, each class of dataset obtained from a field repository folder was uploaded and labelled. Step 3: The “TrainModel” button is clicked to train the model, training starts in Fig. 4.1. This step is achieved within 4 minutes 30 seconds. Step 4: Export the Model. However, neural net models, due to their large file size and data format, cannot run on edge computing devices with resource-constrained memory space and computing power and hence must be scaled down. As a result, the trained module is ported to a microcontroller using the TensorFlow Lite technique that ports the model from the training programming language into a compiled byte array that can be used by an edge processor. This step includes ensuring that the compiled model will fit on the small flash memory footprint and can execute on the available SRAM. The compiled model is ported to the MCU for live testing on onscreen data in real-time. Testing Module of the TeSLF Embedded System Below are the steps involved in setting up the system for operation: Step 1: Each of the three devices (Device A, B and C) is attached to the sheep’s body just a little behind the right fore-leg and strapped gently for stability. Device A is strapped to the lead male , device B is strapped to the lead female , while device C is strapped to the second female sheep. Figure 13: The TeSLF Embedded System Figure 14: The embedded system on the Sheep Figure 15 Logging into MathWorks Platform Results Figure 16: The Embedded System’s Live Data Captured from the Farm Table 1: Real-Time Detection Results from Smart Model (displayed in part) Time Whether Parameter Sheep Body Temp. Degree Sheep Body We
