Vegetation Detection Using UAV Imagery with Deep Learning Segmentation Models

Agricultural fields can be imaged with highresolution multispectral imaging to identify minute changes in plant health and stress indicators before obvious symptoms appear. In agricultural regions, UAVs fitted with multispectral sensors are frequently utilized for mapping and vegetation detection. RGB and near-infrared (NIR) bands are necessary for vegetation detection by UAVs, which despite the fact that multispectral camera UAVs can be more costly to acquire. In this paper, we investigate the effectiveness of pre-trained conventional deep learning models for vegetation detection. Using a dataset of UAV crop imaging, the detection performance of the UNet, DenseNet, and DeepLabv+3 models was evaluated. The results indicate that the U-Net model performed more accurate and visually consistent semantic segmentation's, having 83.72% accuracy which is considerably better then other models.