Improving segmentation of agricultural patterns in aerial imagery using near-infrared images
Abstract: Climate change and loss of biodiversity are currently two of the most imminent global threats caused in part by agricultural practices. With the rise of computer vision, the use of precision farming has become a viable technique to optimise food production. In this thesis, computer vision is utilised to detect nutrient deficiency and weed clusters in aerial images of farmland using existing annotated data. The effect of using near-infrared data in addition to RGB images is analysed in terms of prediction quality. Furthermore, it is evaluated whether including plant indices computed from the raw data can be used to achieve similar results. Different machine learning models are evaluated to verify that the results were similar regardless of architecture. The findings indicate that the near-infrared data is useful for detection of agricultural patterns, nutrient deficiency in particular. The inclusion of plant indices however, does not improve the results.
AT THIS PAGE YOU CAN DOWNLOAD THE WHOLE ESSAY. (follow the link to the next page)