Maize Tassel Detection From UAV Imagery Using Deep Learning.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Flowering time critically impacts crop productivity and yield.
• Maize flowering is sensitive to environmental stress, significantly affecting harvest outcomes.
• Accurate monitoring of maize flowering stages is essential for optimizing agricultural management.

Purpose of the Study:

• To develop and compare deep learning models for automatic maize tassel detection using UAV imagery.
• To evaluate the performance of a custom tassel detection CNN (TD-CNN) against a Faster R-CNN baseline.
• To establish evaluation criteria relevant to agricultural tassel detection needs.

Main Methods:

• Utilized imagery from unmanned aerial vehicles (UAVs) for data collection.
• Developed a custom Convolutional Neural Network (CNN) framework for tassel detection (TD-CNN).
• Employed the Faster Region-based CNN (Faster R-CNN) as a state-of-the-art baseline for comparison.

Main Results:

• Both TD-CNN and Faster R-CNN demonstrated high accuracy in detecting maize tassels.
• The TD-CNN achieved an F1 score of 95.9%.
• The Faster R-CNN achieved a superior F1 score of 97.9%.

Conclusions:

• Deep learning models show significant promise for accurate aerial-based tassel detection in agriculture.
• Further investigation into CNN architectures can enhance accuracy, speed, and generalizability.
• Automated tassel detection can aid in optimizing crop management and yield prediction.