Cotton Yield Prediction via UAV-Based Cotton Boll Image Segmentation Using YOLO Model and Segment Anything Model (SAM)

Abstract

Accurate cotton yield prediction is essential for optimizing agricultural practices, improving storage management, and efficiently utilizing resources like fertilizers and water, ultimately benefiting farmers economically. Traditional yield estimation methods, such as field sampling and cotton weighing, are time-consuming and labor intensive. Emerging technologies provide a solution by offering farmers advanced forecasting tools that can significantly enhance production efficiency. In this study, the authors employ segmentation techniques on cotton crops collected using unmanned aerial vehicles (UAVs) to predict yield. The authors apply Segment Anything Model (SAM) for semantic segmentation, combined with You Only Look Once (YOLO) object detection, to enhance the cotton yield prediction model performance. By correlating segmentation outputs with yield data, we implement a linear regression model to predict yield, achieving an R2 value of 0.913, indicating the model’s reliability. This approach offers a robust framework for cotton yield prediction, significantly improving accuracy and supporting more informed decision-making in agriculture.

Keywords

• cotton
• semantic segmentation
• unmanned aerial vehicles
• segment anything
• yield estimation
• object detection