Cotton Yield Estimation From Aerial Imagery Using Machine Learning Approaches.

Area of Science:

• Agricultural Science
• Remote Sensing
• Machine Learning

Background:

• Pre-harvest cotton yield estimation is crucial for breeding programs, researchers, and producers.
• Traditional field measurements are labor-intensive and less consistent than remote sensing.
• Aerial imagery offers an efficient alternative for crop monitoring and yield prediction.

Purpose of the Study:

• To develop a data processing pipeline for rapid and accurate pre-harvest cotton yield prediction.
• To utilize machine learning techniques with aerial imagery for cotton breeding fields.
• To establish a reliable method for estimating cotton yields from plot-level imagery.

Main Methods:

• A Support Vector Machine (SVM) classifier was trained using selected features to identify cotton pixels from plot images.
• Morphological image processing and a connected components algorithm were applied to classified cotton pixels.
• Cotton boll counts were predicted at the plot level by clustering the processed pixels.

Main Results:

• The SVM classifier achieved 89% accuracy in recognizing cotton pixels.
• The developed model demonstrated a strong fit with ground truth counts, showing an R-squared value of 0.93.
• The model achieved a normalized root mean squared error of 0.07 and a mean absolute percentage error of 13.7%.

Conclusions:

• Aerial imagery combined with machine learning provides a reliable and efficient method for pre-harvest cotton yield prediction.
• The developed data processing pipeline enables fast and accurate yield estimations.
• This approach supports advancements in cotton breeding programs and agricultural management.