Hyperspectral Imaging Combined with a Deep Residual Network for a Precision Nutrition Prediction Model in Commercial Laying Hens

Abstract

The aim of this study was to develop a precision nutrition prediction model for commercial laying hens based on hyperspectral imaging technology and a deep residual network. Eggs from a large‑scale laying hen farm (20,000 birds) were used as research subjects. Hyperspectral image data in the 400–1000 nm range were collected, and the contents of crude protein, crude fat, lecithin, and cholesterol were determined by conventional chemical analysis. An improved deep residual network (ResNet‑SP) was proposed for spectral feature extraction and nutritional index prediction, and its performance was compared with partial least squares regression (PLSR), support vector machine (SVM), and the conventional residual network (ResNet50). The results showed that the ResNet‑SP model achieved the best prediction performance for the four nutritional indices. For crude protein content, the coefficient of determination for the prediction set (R²_p) was 0.931, the root mean square error of prediction (RMSEP) was 0.87 g/100 g, and the residual predictive deviation (RPD) was 3.82. This study verifies the feasibility of combining hyperspectral imaging with deep residual networks for non‑destructive detection of nutritional quality in eggs, providing technical support for precision nutrition management in large‑scale farms.