Abstract: To address the challenge of identifying early-stage subcutaneous hidden injuries in Red Fuji apples induced by mechanical damage—injuries undetectable by the naked eye or RGB imaging—this study proposes a nondestructive detection method for such latent damage based on transmission hyperspectral technology. A transmission hyperspectral system was used to collect spectral data in the 400–1000nm wavelength range from the regions of interest (ROIs) of apples, including areas with latent mechanical damage and healthy, undamaged tissue. The raw spectral data were subjected to five preprocessing procedures including baseline correction (BC), following which feature wavelength selection was conducted via the successive projections algorithm (SPA) and competitive adaptive reweighted sampling (CARS). Principal component analysis (PCA) was combined with minimum noise fraction (MNF) transformation to reduce the dimensionality of hyperspectral images at the selected feature wavelengths. Threshold segmentation and morphological algorithms were then applied to accurately segment subcutaneous mechanical damage areas, with the full-pixel spectral information of the damaged ROIs extracted for subsequent model training. Support vector machine (SVM) and backpropagation (BP) neural network models for identifying subcutaneous latent mechanical damage in Red Fuji apples were established, trained on full-waveband and feature-selected spectral data, respectively. The results show that BC is the optimal preprocessing method for mechanical damage detection in transmission mode, achieving an overall accuracy of 81.67%—an increase of 8.9 percentage points compared with the original spectral model. The first principal component MN1 and the third principal component MN3 images at the 709nm feature wavelength effectively segmented subcutaneous latent mechanical damage areas in the apple samples. Among all the constructed models, the BC-CARS-BP prediction model exhibited the best performance, attaining an accuracy of 97.50% on the prediction set and enabling effective discrimination between healthy apple tissue and early-stage subcutaneous latent mechanical damage of varying degrees. This study confirms that transmission hyperspectral imaging technology can precisely capture the optical characteristics of early-stage, non-browning subcutaneous bruising in Red Fuji apples, thereby providing technical support and methodological references for the nondestructive detection of early mechanical damage during postharvest storage and transportation of apples.