Cotton Yield Estimation Based on Vegetation Indices and Texture Features Derived From RGB Image

Area of Science:

• Agricultural Science
• Remote Sensing
• Precision Agriculture

Background:

• Accurate cotton yield monitoring is crucial for agricultural productivity.
• Unmanned aerial vehicle (UAV) remote sensing offers efficient data acquisition for crop assessment.
• Visible vegetation indices have limitations in differentiating cotton from mulch film and can experience saturation.

Purpose of the Study:

• To investigate the potential of UAV-based visible vegetation indices and texture features for cotton yield monitoring.
• To develop and compare models for cotton yield estimation using these remote sensing data.
• To address the limitations of traditional vegetation index-based methods.

Main Methods:

• RGB images of cotton canopy were captured using a UAV before harvest.
• Visible vegetation indices and texture features were extracted from the UAV-acquired RGB images.
• Linear and nonlinear models, including RF-ELM, were developed using selected features to estimate cotton yield.

Main Results:

• Both vegetation indices and texture features extracted from ultra-high-resolution UAV imagery showed significant correlation with cotton yield.
• The combined vegetation indices and texture features in the RF-ELM model achieved the highest accuracy, with R² of 0.9109 and RMSE of 0.91277 t/ha.
• The model demonstrated a relative RMSE (rRMSE) of 29.34%.

Conclusions:

• UAVs equipped with RGB sensors show significant potential for accurate and non-destructive cotton yield monitoring.
• Integrating visible vegetation indices with texture features enhances yield estimation accuracy, overcoming limitations of individual methods.
• This research provides a valuable theoretical basis and technical support for field cotton production evaluation and management.