GridFree

GridFree automates image-based high-throughput phenotyping of grain kernels to segment, count, and measure kernel traits such as length, width, area, and thousand-kernel weight for genetic and yield studies.

Key Features:

• Python implementation: Implemented as a Python-based software package for image analysis workflows.
• Unsupervised learning: Uses K-Means clustering combined with principal component analysis (PCA) for image segmentation.
• Multi-channel and color-index analysis: Applies PCA and clustering to raw image channels and derived color indices to separate kernels from complex backgrounds and noise artifacts.
• Divide-and-combine strategy: Employs a dynamic threshold mechanism within a divide-and-combine approach to address merged or adjacent kernels.
• Adaptive splitting and merging: Detects outliers in distribution plots of kernel area, length, and width to inform adaptive splitting and merging of segmented objects.
• Quantitative measurements: Extracts kernel-level measurements including area, length, width, and count, and supports calculation of thousand-kernel weight.
• Optional scaling: Provides an optional scaling procedure using a reference object to ensure measurement accuracy across scales.
• Benchmark evaluation: Demonstrated smallest error margins in seed counting compared to existing phenotyping software for crops including alfalfa, canola, lentil, wheat, chickpea, and soybean.

Scientific Applications:

• High-throughput grain phenotyping: Automated segmentation and measurement of kernel traits for large-scale phenotyping experiments.
• Genetic and yield studies: Provides quantitative trait data (length, width, area, thousand-kernel weight, count) for genetic analysis and crop yield improvement research.
• Cross-crop benchmarking: Seed counting and measurement across diverse crops such as alfalfa, canola, lentil, wheat, chickpea, and soybean for comparative studies.

Methodology:

Performs PCA on raw image channels and color indices followed by K-Means clustering for segmentation; applies a divide-and-combine strategy with dynamic thresholds and adaptive splitting/merging informed by outliers in kernel area, length, and width distributions; computes kernel area, length, width, count, and supports optional scaling using a reference object.