Farmers would like to be able to analyse high resolution drone derived imagery to obtain practical information about their crops, such as crop yield, height and emergence. Furthermore, disease monitoring and weeds control could also be other relevant applications. The advent of deep learning and more specifically Convolutional Neural Networks (CNNs) made crop detection on spatial data more affordable. Today different data driven approaches are applied to precision farming improving the ability to assess the crop field status.
Recently advanced algorithms such as Mask RCNN and Retinanet for crop detection and delineation have shown to perform good. However, these algorithms do not use spatial knowledge of for example the crop planting distance or crop rows to distinguish between crops and weeds. Therefore, is important to cope deep learning models with spatial relationships between the target objects. One promising approach are Capsule Networks.
This thesis consists of the following activities: 1) Literature study to find potential spatial deep learning approaches, 2) Implement this approach in Python with a deep learning library of choice, e.g. Keras, Pytorch, 3) Apply the method in the orchards weed detection case study.
Andreas Kamilaris, Francesc X. Prenafeta-Boldú, Deep learning in agriculture: A survey, Computers and Electronics in Agriculture, Vol. 147, pp. 70-90, 2018.
He, Kaiming et al. “Mask R-CNN.” 2017 IEEE International Conference on Computer Vision (ICCV) (2017)
- UAV enthusiast
Willing for learning cutting edge machine learning algorithms
Excited to solve problems that will help people
Theme(s): Sensing & measuring