Cell phone image analysis as a proxy for laboratory egg quality measurements.

Precise individual-level recording of egg quality for genetic improvement requires laboratory instruments often unavailable at field test locations or in developing countries where breeding programs where local breeds are being developed. This study evaluated smartphone camera-derived measurements validating them against laboratory references for shell color (measured with Minolta colorimeter), egg weight, and defect classification in 359 matched eggs (180 brown, 179 white; 180 defective, 179 normal). Eggs were segmented out using SAM3 text-prompted segmentation applied to each raw photograph. A custom Python image pipeline extracted CIE L* (lightness), a*(redness on red-green scale), b*(yellowness on blue-yellow scale), chroma, morphometric axes, projected area, shape index, and surface defect scores from a mean of 2 images per egg acquired with a smartphone stabilized on a tripod, additionally a scale calibration from a physical reference sticker was considered. Combined-group validation statistics for image-derived color components against colorimeter values were high (Pearson r ≥ 0.986, 10-fold cross-validated R2 ≥ 0.971) but inflated by the Brown/White group contrast, which explained ≥ 95% of total variance in every color component. Egg weight was predicted from pixel-based morphometric measurements (egg_area_px + long_axis_px + short_axis_px) with 10-fold cross-validated R2 = 0.922 overall, 0.961 for Brown eggs, and 0.842 for White eggs (cross-validated RMSE = 1.149 g). Shell breaking strength was weakly predicted by image features (cross-validated R2 ≈ 0.11). A soft-voting ensemble (Random Forest + Gradient Boosting + SVM) trained on 30 SAM3-derived image features achieved AUC = 0.810 ± 0.078, accuracy = 0.730 ± 0.052, sensitivity = 0.744, and specificity = 0.715 for binary defect classification; a decision tree and logistic regression on 13 summary features each achieved AUC = 0.673 by the same 10-fold cross-validation. These results demonstrate that smartphone-based image analysis accurately replaces laboratory colorimetry and weighing, shifting the operational challenge from instrument access to data transfer and computational analysis, and thereby extending the reach of genetic improvement programs.