Cotton Stand Count Using UAS Imagery and Deep Learning

Stand count is critical for growers to make decisions for replanting and other site-specific management to avoid yield loss. Manual count in the field is time consuming and labor intensive. With high-resolution UAS imagery, alternative approaches are proposed in this study. One approach is to use a simple image processing procedure to locate and count the cotton stand in RGB images. Another approach is to use deep learning Faster Region-based Convolutional Networks (RCNN) and CenterNet. Faster-RCNN is a simple and basic object detection algorithm, while CenterNet is a state-of-the-art object detection model. The proposed approaches were tested with different UAS datasets containing variability in plant size, cultivar and background brightness. We evaluated the overall performance with different inputs and cost with UAS imagery in cotton stand count using the simple image-processing based approach and the advanced deep learning algorithms.