Machine learning in agriculture: 100% savings on pesticides and human resources

Innowise incorporated computer vision technology into self-driving agricultural robots, enabling them to automatically feed plants and remove weeds with lasers.

Customer

Industry

Agriculture, IoT

Region

Client since

2021

Our client is a company producing autonomous agricultural robots to automate and accelerate farm work within the European region.

Detailed information about the client cannot be disclosed under the provisions of the NDA.

Challenge: Overcoming limitations of manual plant care with AI technology in agriculture

The utilization of ML farm systems and robots in the agricultural sector is becoming increasingly crucial due to the significant hurdles posed by manual plant care, which demands extensive human effort, time, and expenses. These advanced technologies can address various challenges, including labor scarcity and resource efficiency. This results in a more comprehensive and efficient solution to modern agriculture issues.

Our client produces autonomous robots and devices that are supposed to automate the process of cultivating and nurturing plants. Although the robots could move around the beds and fields, they lacked the ability to differentiate between plants and weeds for selective fertilization and watering purposes.

Our experts were faced with a significant challenge of integrating specialized software into the robots that could precisely distinguish and segregate thinned-out plants. The subsequent objective for the program was to eliminate specific weeds using lasers with optimal accuracy. Furthermore, ML farm systems needed to determine the type of plants and supply them with a sufficient amount of suitable fertilizer, depending on their class and condition metrics.

In summary, the scope of work included:

data collection;
manual data markup;
data augmentation;
model training;
model integration;
real-time processing.

Solution: Implementing machine learning in the agricultural sector for weed elimination and selective plant feeding

Our developers successfully completed the project and implemented an AI solution into the autonomous robots for the real-time processing of scanned field images and identifying weeds in milliseconds. Equipped with precisely calibrated lasers, the robots can eliminate up to 100,000 weeds per hour. Additionally, the robots are now capable of classifying plants and administering fertilizers based on their individual requirements. Furthermore, they can determine field conditions and metrics to optimize agricultural practices and enhance efficiency.

Utilizing an integrated video camera, we gathered and labeled a dataset consisting of more than 10,000 plant images. Our team then proceeded to perform tasks such as marking, augmentation, and model training on an expanded dataset. Innowise has successfully implemented a supervised machine learning model that can automatically establish the connection between input variables and target outputs, enabling precise predictions on novel, unseen stem and field images. This facilitates further plant classification and stem detection, weed eradication, and selective fertilization. This solution serves as an outstanding use case of machine learning in the agricultural sector, delivering remarkable outcomes in terms of automating tasks, conserving resources, enhancing fertility, and mitigating the adverse environmental impact caused by pesticides.

End-to-end plants segmentation and stem detection neural network

During the data acquisition phase, we gathered plant and weed images through a video camera attached to an agricultural robot navigating a field. Once acquired, agricultural specialists marked up the data for object detection and segmentation in subsequent stages of data augmentation and refinement.

Subsequently, our team developed a custom neural network capable of identifying the type and class of a plant from an image and making informed decisions on plant treatment based on prior experience. We integrated this solution into an end device equipped with GPU, allowing it to process real-time data and distinguish plants from previously learned datasets. The stem detector identifies plant stem locations to facilitate laser guidance.

The software enables decision-making by the robot without Internet access while working in agricultural fields. Upon returning to the station and accessing the network, the dataset can be updated with additional information and settings. The neural network’s capabilities are not confined to one database: the machine learning system supports the retraining of neural networks using updated datasets to grow new plant types and eradicate various types of weeds.

Aside from identifying plant and weed classes, the neural network can also ascertain the field’s condition and key metrics, subsequently used to regulate irrigation intensity.

High-precision laser weed elimination and selective plant feeding

The ML farm systems utilize cutting-edge technology to revolutionize the agricultural industry. During the data acquisition phase, the integrated video camera collects plant and weed images as the agricultural robot moves through the field. The collected data is then marked up by agricultural specialists for subsequent object detection and segmentation.

The end-to-end crops and weeds segmentation neural network provides accurate semantic segmentation of the scene, distinguishing crops, weeds, and grass. The system sends signals to several laser modules that operate simultaneously, allowing the autonomous weeders to kill over 100k weeds per hour, automatically and chemical-free. The laser system’s high accuracy is due to ultra-precise detectors, with finely tuned laser parameters making it possible to determine the range up to 2 mm.

The system also employs selective feeding, which treats each plant on the field individually. The computer vision analyzes each plant’s current state, considering factors such as growth stage, health status, and nutrient requirements. Based on this information, the system determines the most appropriate treatment for each plant, selecting the right feed portions to apply. This leads to a reduction in resources and a more cost-effective approach to plant feeding.

The ML farm systems are designed to be flexible and adaptable to various types of plants. The neural network can learn and relearn from new data sets, which can be used to train the AI engine to identify and treat different species of plants. This involves collecting and labeling images of the new plants, performing data augmentation and refining the new data, allowing the system to continually expand its knowledge base and capabilities.

Overall, the ML farm system developed by Innowise is an excellent example of the benefits of machine learning in the agricultural sector, enabling cost-effective and efficient solutions for crop management and treatment.

Technologies

ML & MLOps

Python, PyTorch, OpenCV, MMSegmentation, TensorFlow, AWS (S3, Lambda, EC2, CloudWatch)

Process

Our team conducted an initial meeting with the client to gather requirements and understand their specific needs for the autonomous robots. Based on these requirements, we created a comprehensive design plan for developing the software system, which consisted of two main stages: data collection and labeling using an integrated video camera and the implementation of a supervised machine learning model.

To manage the project effectively, we followed the Agile methodology and held daily meetings to track progress and discuss any issues or concerns. We also utilized communication tools like Google Chat and project management software like Jira and Confluence to assign tasks and monitor performance.

After a month and a half of development, we were able to create the MVP version of the neural network, which was capable of making effective decisions without additional control. This approach allowed us to develop a flexible and scalable system that could be adapted to different agricultural settings and use cases, providing farmers with a cost-effective and efficient solution for managing their operations.

Team

Project Manager

ML Engineers

Back-End Developer

Results: Reduction in pesticide usage and human resource costs

The implementation of machine learning in agriculture through the use of agricultural robots equipped with computer vision and AI-based engines provides numerous benefits for the industry. It promotes cost-effectiveness by reducing the use of unnecessary fertilizers and chemicals and improving agricultural productivity through selective treatment of each plant. Moreover, it offers detailed field monitoring and mapping without human intervention, providing farmers with vital information on their fields’ condition.

The result of implementing this technology for the client is a reduction in overall resources used, leading to economic benefits through continuous automatic crop care, high yields, and perfect plant health. Additionally, laser-based, chemical-free weed elimination protects agricultural ecosystems, minimizing the negative environmental impact of traditional farming practices. The system’s ability to continuously learn and adapt allows farmers to update the data set regularly and adapt to new types of plants and agricultural work.

Overall, the integration of AI technology in agriculture has enormous potential to bring benefits to the industry, the environment, and nature. ML robots can increase crop quality and fertility, reduce costs, preserve natural resources, and eliminate potential harm to humans by completing complex tasks automatically.

Project duration

September 2021 - November 2022

64%

fertilizer savings

100%

savings on pesticides and human resources

Related cases

Contact us!

Book a call or fill out the form below and we’ll get back to you once we’ve processed your request.

What happens next?

Having received and processed your request, we will get back to you shortly to detail your project needs and sign an NDA to ensure the confidentiality of information.

After examining requirements, our analysts and developers devise a project proposal with the scope of works, team size, time, and cost estimates.

We arrange a meeting with you to discuss the offer and come to an agreement.

We sign a contract and start working on your project as quickly as possible.