PhD project: Using unmanned aerial systems and machine learning for precision agriculture

Name: Pornntiwa Pawara

Supervisors:
Prof. dr. L.R.B. (Lambert) Schomaker
Ddr. M.A. (Marco) Wiering

Summary of PhD project:

In 2050, the world population is estimated to be over 9 billion people. For feeding all these people, technical advances need to be used in agriculture, which constitutes the field of precision agriculture. In this research project we will make use of unmanned aerial systems (drones) to monitor crops on the land in order to assist a human farmer. Using the cameras of the unmanned aerial systems we want to answer the following research questions: which visual patterns can be detected on crops that are most useful to help a farmer in his/her daily practice? Which methods from machine learning and computer vision are best in detecting and classifying these patterns?

To answer these questions, the research will cover a wide variety of computer vision, pattern recognition, and machine learning techniques, in order to develop a novel recognition system that can detect and classify different aerial images of plants. In the project the unmanned aerial system (UAS) will be used to fly over specific farming lands containing different types of plants. During the flight the bottom camera mounted on the UAS will be used to automatically collect a large amount of images. The UAS can first fly at a higher altitude to have a broad view on the land. After that the recognition system should learn which parts of the land deserve a closer look, for example because of possible sick or dry plants in a particular area. Then the UAS will fly over these areas of interest at a lower altitude and collect a large number of images. From these images, the system can report to the farmer if there are particular plants, which deserve specific attention from the farmer.

The recognition system should be continuously learning as new data (images, plant labels) arrive. Furthermore, we do not want to focus on one type of crop or images taken at specific times/periods the system should be generally applicable to a wide variety of images of crops, because it can be trained with visual patterns and specific labels. The recognition system will be combined with particular tools to map the land and show complete images to the participating farmers. The project could lead to a new generation of quite cheap techniques that can help farmers to increase their production.