“”

Success Story: AI-powered Farming Automation

3 min read

Farming automation is hitting record-breaking levels. According to the most recent Western Growers report, about 70% of growers invested in automation last year. Harvest assisting and weeding robots were the top priorities of the field automation process.

As a technology company, we don’t remain on the sidelines! This post reviews our contribution to the farming automation project that made a sound technological impact.

Building the AI Farming Automation System for the agricultural holding

In this project, we performed as a development and technology partner for an Asian agricultural holding that intended to cut expenditures and maximize cow productivity. We developed customized hardware and software engineering solutions to address these goals. Our approach was developed for large-scale farming as the partnering holding operated a herd of 3000 cows and three dairy farms.

Our farming automation concept implied building a one-stop system that fully satisfies the project goals set by the client. We approached this concept through Artificial Intelligence and Machine Learning, relying on IoT systems (sensors) for data collection. First, we sought to understand farming activity with the help of data.

Challenge

We intended to collect the raw data with the help of sensors. Our goal was to establish control over the condition and behavior of cows by configuring IoT systems and synchronizing their operation.

However, no manufacturer on the market could fully match our sensor requirements. Of course, there is no shortage of hardware products in the world. But the security considerations did not allow us to choose hardware centered around cloud storage. Also, we were stipulated by technical limitations of equipment that could be employed on a large scale in the AI-backed Farming Management System.

That’s why we decided to conduct deep modernization and customization of equipment components supplied by different manufacturers.

Solution

An intense R&D phase allowed us to understand how equipment behaves in various conditions. Farm automation equipment included multi-channel oscilloscopes, spectrum analyzers, multi-spectral cameras, IR and UV light sources, thermal imagers, and pyrometers.

The hardware equipment was divided into groups:

  1. Wearable animal sensors;
  2. Wearable sensors for staff;
  3. Stationary sensors on farms;
  4. Data transmission equipment;
  5. Equipment for collecting and storage of information.
jakob ben cotton K1hwkV5GPl0 unsplash development
Automation project for the cow farm

 

Our data collection systems worked permanently, regardless of whether an animal was on the farm or field. Data was sent into storage in two ways:

a) from on-farm stationary sensors in batches every 10 minutes;
b) from on-animal sensors systems when animals were grazing in the field — the collected data is automatically reset when the herd returns to the farm.

We collected the following list of parameters that should be controlled:

  • the walking distance and speed
  • the amount of time spent in one place
  • a record of eating and drinking facts that allow for spotting deviations in the process of eating (rumination)
  • geo-position
  • the position of the animal body and its parts in space
  • convulsions
  • basic vital signs
  • position relative to other animals
  • gait disturbance
  • milk parameters
  • whether the milking and feeding routines are followed
  • milk productivity
  • indoor and outdoor microclimate
  • staff duties schedule
  • livestock morbidity
  • pregnancy characteristics.

Having studied similar solutions existing on the market, we came up with the following stacks:

  1. Equipment that monitors animals’ behavior: a micro radar for monitoring heart rate, 6-axis accelerometers mounted on the collar from different sides, electret microphones, geo-position tracker, Wi-Fi tracker for precise indoor position control, RFID tag, several contact temperature sensors, controller for collection, storage, and transmission of information.
  2. Employee monitoring sensors: we developed a compact personal device with an in-built movement tracker, a microphone, an RFID tag, and a controller for storing and transmitting information
  3. Stationary equipment: video cameras, methane and carbon dioxide gas analyzers, temperature, humidity, illumination, wind speed, and direction sensors, remote thermal sensors (pyrometers), sensors for measuring levels (forage, water, manure, etc.), network equipment for data transmission to the server.

We chose Hikvision HD cameras with IR illumination to solve the problem of data extraction at night, which is essential for monitoring the birth and behavior of young animals.

Data processing for farming automation

1) Following the data collection, analytical modules systematized information on production processes, cow feeding, care routines, staff responsibilities, microclimate parameters, and genealogical and veterinary data. Our Data engineers created data preprocessing and validation pipelines based on Apache Kafka.
At this step, we applied the following stacks: Spark, Databricks, PostgreSQL, and S3.

2) Training of the neural network based on the data. This implied data standardization, filling in missing values, and splitting the data into train and test datasets. The applied stacks were Pandas and Dbt.

3) Next step was to create and train a neural network using ML algorithms such as MLP (Multilayer Perceptron) or SVM (Support Vector Machine). Neural networks were trained on a dataset of 1500 images and 17 GB of text and tabular data. We used a “lightweight” model of the YOLO-v4 detector – YOLO-v4_tiny. We applied non-parametric and robust regression for tabular data using a t-model. The applied stacks: TensorFlow, PyTorch, Python, C++, Git, Keras, Stan.

4) It was necessary to evaluate the model quality using accuracy, standard deviation, and coefficient of determination metrics. If the model did not meet the requirements, we made additional adjustments or changed the parameters of the training algorithm. Stack: Mlflow

You can compare the “before and after” dynamics:

Screenshot 3 development

Project impact

Successful integration of AI&ML-powered monitoring systems became a game changer for the Asian agricultural holding:

  1. Milk production at the dairy farms was increased by 28-30%
  2. The holding cut its expenses by 20-22%

    12.07 Solving Agricultural Challenges with AI and ML Solutions img11 development
    AI Farming automation

     

Bottom line

We help integrate AI and ML technologies into farm management systems that consolidate data from various sources, such as sensors, machinery, and weather stations. These systems provide farmers with comprehensive insights into their operations, facilitating data-driven decision-making and optimizing overall farm performance.

Meanwhile, you can check our previous reviews of smart agritech solutions and subscribe to our blog to keep up with further publications!

Want to discuss your project? Let's talk!
By submitting request you agree to our Privacy Policy

Editor's Choice

Post Image
8 min read

Turn Your Design System into a Massive Hit!

When you have a large-scale business, your team can face numerous challenges while creating user experiences. One solution to these challenges is the…

Post Image
8 min read

The Secret to Top-Notch Software Development: Our HR Management Platform

Software development is knowledge-driven and labor-intensive, meaning our most valuable assets are our employees. We must hire and retain the best teams to…

Post Image
6 min read

Software Engineering Culture and How we in JetRuby Develop It

Have you ever wondered what a software engineering culture is? What if we revealed that it’s one of the primary reasons your clients…

Get the best content once a month!

Once a month you will receive the most important information on implementing your ideas, evaluating opportunities, and choosing the best solutions! Subscribe

Contact us

By submitting request you agree to our Privacy Policy