A Deep Learning Approach to Estimate Plant Density and Spread: Application to Durum Wheat and Sorghum Crops
Abstract
Plant counting and spatial distribution of emergence is important factors that are related to many aspects of wheat crop productivity including emergence rate, seed quality, overall plant performance, and yield. Manual plant counting in the field is time consuming and requires substantial human efforts. Most studies on plant counting and density estimation using remote sensing have focused on emergence count and stand density. However, spread of plant stands and spacing plays an important role in plant performance and yield. It is imperative to develop alternative techniques to automate the process of plant density estimation from imagery. Deep learning is an emerging field in realm of machine learning that provides unprecedented performance on wide ranges of data analysis problems, and it can be used to automate the manual methods with high accuracy and efficiency. In this study, we employ a deep learning approach, namely Mask R-CNN, for plant density, area and spatial distribution estimation of Durum wheat and sorghum crops from more than 100 genotypes of each crop in different growth stages and image spatial resolution. The remote sensing images captured by a field-robot, the largest robotic field scanner in the world, are used to automatically compute plant density and distribution in near real-time. In addition, analysis of plants in different spatial resolution allows for a characterization of plant density and can be implication to multiple imaging platforms such as unmanned aerial systems and satellites. The results of this study demonstrate the potential of low-cost sensor supported by a field robot and a new deep learning technique can provide accurate estimation of plant traits.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFMGC23G1416Y
- Keywords:
-
- 0402 Agricultural systems;
- BIOGEOSCIENCES;
- 1630 Impacts of global change;
- GLOBAL CHANGE;
- 1632 Land cover change;
- GLOBAL CHANGE;
- 1640 Remote sensing;
- GLOBAL CHANGE