Open AccessDevelopment of a small scale cartesian coordinate farming robot with deep learning based weed detection
Author(s) -
T. Rajalakshmi,
Patrick Panikulam,
Patil Kshunotra Sharad,
Rajeev Ravindran Nair
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1969/1/012007
Subject(s) - cartesian coordinate system , automation , computer science , artificial intelligence , agricultural engineering , precision agriculture , scalability , real time computing , simulation , agriculture , mathematics , engineering , geography , database , geometry , mechanical engineering , archaeology
Automated Cartesian coordinated farming is a system designed for agricultural purposes. Being one of the trends of development on automation and intelligence in the agricultural machinery, this system is able to perform certain basic elementary functions like seed sowing, spraying, watering, etc. The idea of robotics technology is being applied in agriculture. This is being designed in minimizing the labor of farmers apart from increasing the speed and accuracy of the work. A scalable Cartesian coordinate based system is modeled, which can take care of a particular area of farm land or a garden until it is time for harvesting. The system starts by planting individual seeds at predetermined locations and then automatically waters it with the exact amount required for each type of plant. It has the ability to measure soil humidity and rainfall so that water can be used depending upon on the nature of the day. The proposed system makes use of a YOLO (You Only Look Once) object detection technique to detect weeds. YOLO processes plant images at 45 frames per second in real-time, which is faster than other object detection techniques. Here, the image is divided into several grid cells before being processed. The bounding boxes as well as the class probabilities are predicted by one single neural network, in a single evaluation. This effectively boosts the speed and accuracy of weed detection.