MASK R-CNN APPROACH TO DETECT HEALTHY VEGETATION AREAS IN NIR IMAGES

Resumo

With the development of new technologies in the field of deep learning, sectors such as agri-
culture have been benefited from the application of intelligent systems allied to the use of UAVs (Un-
manned Aerial Vehicles) in crop monitoring to quickly and accurately detect specific areas of vegetation

and optimize decision making to ensure the quality of planting. Some researchers inspired from the
deep learning advance and its success in many areas have been studied solutions in detection of healthy
vegetation areas, but, how better are the performances of advanced techniques compared to traditional
techniques? In this context, this paper presents a comparison between a traditional technique (K-Means
Clustering) and advanced technique (Mask R-CNN) applied to detect different vegetation areas in NIR

images. The database of this work consists of NIR images provided by a modified RGB camera in-
stalled in a UAV. Basically, as an input were used NDVI (Normalized Difference Vegetation Index), an

important index of vegetation healthy, obtained from the NIR images. Finally, a comparison between the
proposed algorithms for detection of healthy vegetation areas is presented, showing the improvements of
the proposed Mask R-CNN algorithm.