DeepCount: In-Field Automatic Quantification of Wheat Spikes Using Simple Linear Iterative Clustering and Deep...

Area of Science:

• Agricultural Science
• Computer Science
• Plant Breeding

Background:

• Accurate crop yield assessment is crucial for agricultural research and breeding.
• Manual wheat ear counting is labor-intensive, time-consuming, and costly.
• Automated systems are needed for efficient and high-throughput crop analysis.

Purpose of the Study:

• To develop a computationally efficient system, DeepCount, for automatic wheat spike counting in field images.
• To leverage deep learning for robust wheat ear quantification under diverse conditions.
• To provide a foundation for portable, smartphone-assisted, and UAV-based crop monitoring systems.

Main Methods:

• Image segmentation using Simple Linear Iterative Clustering (SLIC) to derive superpixels.
• Feature extraction and model construction for deep convolutional neural network (CNN) classification.
• Semantic segmentation of wheat spikes using a deep learning approach.

Main Results:

• DeepCount demonstrates high robustness across different growth stages, environmental conditions, and canopy complexities.
• The system achieves accurate wheat ear quantification validated against ground-based measurements.
• Performance is superior to traditional edge detection and morphological analysis methods.

Conclusions:

• DeepCount offers a feasible and robust solution for automated wheat spike counting in real-world agricultural settings.
• The system significantly reduces manual labor and supports high-throughput phenotyping.
• Potential for adaptation to RGB images from unmanned aerial vehicles (UAVs) for broader applications.