Click on “Download PDF” for the PDF version or on the title for the HTML version.


If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.

Leveraging transfer learning in ArcGIS Pro to detect “doubles” in a sunflower field

Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan www.asabe.org

Citation:  2021 ASABE Annual International Virtual Meeting  2100742.(doi:10.13031/aim.202100742)
Authors:   N. Rai, P. Flores
Keywords:   Sunflower doubles, Transfer learning, UASs, Double detection

Abstract.

The occurrence of doubles or multiple plant cluster in a sunflower (Helianthus annuus L.) field is undesirable since they can contribute to disease spread leading to yield reduction. A common approach to identify doubles in a sunflower field is by in-person field scouting, which can be arduous and time-consuming task. Therefore, a method was proposed to use unmanned aerial systems (UASs) imagery coupled with transfer learning in ArcGIS Pro to automate the process of identifying and mapping the occurrence of sunflower doubles across the field. The imagery was collected from a sunflower field, around a week after germination using a Phantom 4 Pro flown at an altitude of 40ft above ground level (AGL). The orthomosaic imagery was manipulated in Python where the exported image chips were trained on 4 architectures, namely, Faster R-CNN, RetinaNet, YOLOv3, and SSD. These 4 architectures were further fine-tuned to detect doubles in the sunflower field. Results showed that Faster R-CNN outran the rest of the models to detect sunflower doubles across the whole orthomosaic imagery. The average precision accuracy of Faster R-CNN was 99.6% with an average F1-score of 0.95, precision of 0.91, and recall of 0.96. On the other hand, RetinaNet and YOLOv3 yielded an average precision score of 93.8% and 87.8%, respectively. Based on these scores, we recommend Faster R-CNN to detect doubles on sunflower fields. For our future work, we are working to automate sunflower stand counts while adding another class for weed detection across the field as well.

(Download PDF)    (Export to EndNotes)