Cotton Emergence Uniformity Assessment Using Remote Sensing and Machine Learning

Abstract. Seed emergence uniformity has a significant impact on plant vigor, emergence rate and overall crop health, which ultimately affects fruit production. Identifying the optimal germination rate, taking into account different genotypes and environmental factors, is critical to improving crop yield. This study presents a novel method that uses an unmanned aerial vehicle (UAV) to collect RGB field images and applies deep learning and statistical analysis to evaluate cotton (Gossypium hirsutum L.) uniformity under varying seeding rates. Traditional methods often involve laborious manual plant counting to identify areas of over- or under-emergence that affect lint yield, a time-consuming process for large fields. To address this issue, our study, conducted in southern Missouri, USA, used five different seeding rates. We used a UAV-mounted camera to capture field images two weeks after planting. These images were stitched into an orthomosaic of the entire field and then segmented into smaller blocks. The YOLOv7 object detection algorithm was used to locate each cotton plant within the segmented images. We also used the Hough transform and polynomial regression techniques to identify cotton rows and remove weeds. These methods yielded a mean average accuracy at 50% intersection over the union threshold of 96.8% mAP@50. This study provides valuable insights by developing a pipeline for early-stage cotton stand count and distance estimation using remote sensing techniques. This approach improves the assessment of cotton emergence uniformity, leading to more efficient crop management.