DeepAPS

DeepAPS produces precise segmentation masks from animal images to enable automated phenotyping and extraction of morphological and coat-color measurements for genetic and breeding analyses.

Key Features:

• Composite segmentation: Integrates two existing algorithms into a composite method that generates precise masks for animal images.
• Background removal: Enhances the accuracy and efficiency of background removal from animal photographs.
• Mask-based phenotype extraction: Uses generated masks to extract phenotypic information from multiple morphological features.
• Quantitative measurements: Automatically quantifies up to 14 different morphological phenotypic measurements from masks.
• Partially supervised training: Employs a partially supervised machine-learning approach requiring approximately 50 annotated images for training.
• Validation: Validated against manual classification with an adjusted R² of 0.926 for coat color proportions.
• Application to pedigree data: Applied to pedigree and image data from a web catalog (www.semex.com) to estimate trait heritabilities.
• Heritability estimates: Produced heritability estimates ranging between h² = 0.18 and h² = 0.82 for various traits.

Scientific Applications:

• Automated phenotyping segmentation: Segmentation and background removal for phenotyping in biological and agricultural studies, including dairy industry datasets.
• Trait quantification: Quantification of morphological traits and coat-color proportions for phenotype characterization.
• Genetic and breeding analyses: Extraction of phenotypic data for heritability estimation and genetic studies in breeding programs using pedigree and image data.

Methodology:

Integrates two existing segmentation algorithms into a composite method to generate accurate animal masks, uses masks to extract multiple morphological features, trains a partially supervised machine-learning model with ~50 annotated images, and validates results against manual classification (adjusted R² = 0.926 for coat color proportions).