An Adaptive Partial Least-Squares Regression Approach for Classifying Chicken Egg Fertility by Hyperspectral Imaging

Area of Science:

• Agricultural Science
• Chemometrics
• Biotechnology

Background:

• Partial least-squares (PLS) regression is a predictive modeling technique often used in chemometrics.
• While not initially designed for classification, PLS has shown success in discrimination tasks.
• Non-supervised methods like PCA and k-means are less effective for discrimination compared to PLS.

Purpose of the Study:

• To evaluate the effectiveness of Partial Least-Squares (PLS) regression for discriminating between fertile and non-fertile chicken eggs.
• To apply PLS regression to hyperspectral imaging data for fertility assessment.
• To demonstrate the suitability of an adaptive PLS approach for analyzing chicken egg fertility.

Main Methods:

• Hyperspectral images of chicken eggs (white and brown) were captured in the NIR region (900-1700 nm) on days 0-4 of incubation.
• Spectral information was extracted from a region of interest, and transmission characteristics were averaged.
• A moving-thresholding technique based on PLS regression was implemented for discrimination.

Main Results:

• The PLS technique achieved high true positive rates (TPRs) up to 100% for discriminating fertile from non-fertile eggs.
• Effective discrimination was observed at selected threshold values between 0.50 and 0.85.
• The method demonstrated accuracy across different days of incubation.

Conclusions:

• The proposed Partial Least-Squares (PLS) regression technique accurately discriminates between fertile and non-fertile chicken eggs using hyperspectral imaging.
• The adaptive PLS approach is suitable for hyperspectral imaging-based chicken egg fertility data analysis.
• PLS regression offers a preferable alternative to non-supervised methods for egg fertility discrimination.