Abstract
Based on recent successful applications of Deep Learning techniques in classification, detection and segmentation of plants, we propose an instance segmentation approach that uses a Mask R-CNN model for weeds and crops detection on farmlands. We evaluated our model performance with the MSCOCO average precision metric, contrasting the use of data augmentation techniques. Results obtained show how the model fits very well in this context, opening new opportunities to automated weed control solutions, at larger scales.