Wheat Spike and Spikelet Detection on Close-range Digital Imagery using Deep Learning
Presentation Type
Poster
Student
Yes
Track
Precision Ag/Biological Sciences Application
Abstract
Wheat is one of the primary cereal crops and plays a vital role in global food security, providing essential nutrients and serving as a staple food for billions of people worldwide. Wheat production and supply are facing significant threats due to population growth and climate change. The wheat spike number per unit ground area is a critical factor in wheat production, and monitoring spike count is crucial for maximizing yield potential, ensuring food security, and advancing agricultural practices. The manual identification and counting of wheat spikes and subsequent statistical analysis are labor-intensive and error-prone. The deep learning-based object detection methods using high-resolution digital imagery have demonstrated promising results in automating crop counting, significantly reducing the labor required while maintaining high accuracy. However, the varying size, orientation, shape, and texture of wheat spikes, depending on their variety and growth stage, present a significant challenge. This study uses cutting-edge deep learning-based object detection algorithms like YOLOv11 (You Only Look Once) to accurately detect and count oriented wheat spikes from high-resolution digital imagery collected using drones and mobile cameras. The detected spikes are further used for spikelet detection and counting, providing an end-to-end estimate of spike and spikelet. Our results show that deep learning-based object detection is a rapid and reliable approach for wheat spike and spikelet detection and counting, demonstrating its effectiveness in providing a reliable reference for further wheat yield estimation. The proposed approach is expected to adapt to digital imagery collected from various angles with different spatial resolutions, offering a robust solution for high-resolution wheat spike and spikelet detection and contributing to improved yield estimation.
Start Date
2-7-2025 1:00 PM
End Date
2-7-2025 2:30 PM
Wheat Spike and Spikelet Detection on Close-range Digital Imagery using Deep Learning
Volstorff A
Wheat is one of the primary cereal crops and plays a vital role in global food security, providing essential nutrients and serving as a staple food for billions of people worldwide. Wheat production and supply are facing significant threats due to population growth and climate change. The wheat spike number per unit ground area is a critical factor in wheat production, and monitoring spike count is crucial for maximizing yield potential, ensuring food security, and advancing agricultural practices. The manual identification and counting of wheat spikes and subsequent statistical analysis are labor-intensive and error-prone. The deep learning-based object detection methods using high-resolution digital imagery have demonstrated promising results in automating crop counting, significantly reducing the labor required while maintaining high accuracy. However, the varying size, orientation, shape, and texture of wheat spikes, depending on their variety and growth stage, present a significant challenge. This study uses cutting-edge deep learning-based object detection algorithms like YOLOv11 (You Only Look Once) to accurately detect and count oriented wheat spikes from high-resolution digital imagery collected using drones and mobile cameras. The detected spikes are further used for spikelet detection and counting, providing an end-to-end estimate of spike and spikelet. Our results show that deep learning-based object detection is a rapid and reliable approach for wheat spike and spikelet detection and counting, demonstrating its effectiveness in providing a reliable reference for further wheat yield estimation. The proposed approach is expected to adapt to digital imagery collected from various angles with different spatial resolutions, offering a robust solution for high-resolution wheat spike and spikelet detection and contributing to improved yield estimation.