Crop row identification and plant cluster segmentation for stand count from UAS imagery based on profile and geometry

Abstract

Plant stand count is an important measure to determine the attainment of the target plant population and obtain seed emergence characteristics. Unmanned aerial system (UAS) imagery is generally analyzed with commercial software for estimating plant stand count. However, such software is expensive, crop-specific, and imposes limitations. The available literature, research, and applications typically use such software, whose underlying working principles are unknown. Therefore, a user-coded, open-source, computer vision ImageJ crop row identification and stand count plugin termed “CRISCO” was developed and validated. The plugin integrates profile and geometry-based approaches in a customized framework that perform automatic crop row orientation, row identification, plant cluster segmentation, and plant stand counting from the UAS imagery. The plugin was validated on sunflower field images from two datasets “Set I” and “Set II” representing different flight altitudes, field areas, image resolutions, and growth stages. Crop row identification in the CRISCO plugin accurately identified rows up to

(tested) and it could potentially work with rows even

, provided the rows are straight, which is the case with modern planting methods. The developed segmentation approach by combining profile and geometry termed “ProGeo” resolved the plant clusters. Comparing ProGeo and watershed segmentation, the former produced 89–

of correct segmentation, while the latter produced 51–

. The plant-stand count accuracy of the plugin ranged from 85–

with CPU time for analysis ranging from 2–

for the two datasets. The user-coded plugin, although developed and tested on sunflower, can be extended with appropriate modifications to accommodate other row crops (e.g., soybeans, cotton).