Abstract: To address the difficulty of timely identifying mild bruising in Xinjiang prunes during postharvest sorting and transportation, where internal damage accumulates over time and deteriorates quality, this study developed a Xinjiang prunes mild-bruising recognition and temporal-evolution characterization approach based on reflectance visible and near-infrared spectroscopy (Vis-NIR). First, mildly bruised samples were prepared using a self-designed collision device, and reflectance and transmittance acquisition modes were compared. Reflectance Vis-NIR was selected as the spectral acquisition method. Following a standard white-reference peak control principle and a one-factor-at-a-time strategy, key parameters including integration time, source–sample distance, spectral smoothing, and measurement posture were optimized to ensure spectral repeatability. Next, reflectance spectra were collected at 1, 6, 24, and 48 h after bruising, with intact fruits set as controls. Mean spectral analysis was performed, and dimensionality-reduction visualization using principal component analysis (PCA) and uniform manifold approximation and projection (UMAP) was conducted to characterize spectral differences across time points. Support vector machine (SVM), random forest (RF), and extreme gradient boosting (XGBoost) classifiers were then developed and compared in terms of recognition performance and generalization ability. The results showed that the reflectance of mildly bruised samples changed significantly over time in the 700～950 nm region, with the most pronounced difference from the control at 24 h. Under multiplicative scatter correction (MSC), detrending (DT), and continuous wavelet transform (CWT) preprocessing, the SVM achieved classification accuracies above 92.00%, and the CWT–uninformative variable elimination–SVM (CWT-UVE-SVM) model reached an accuracy of 96.67% at 24 h. A model trained on mildly bruised samples maintained good generalization in transfer validation on moderately and severely bruised samples, achieving 99.17% accuracy for severe bruising. In addition, incorporating a time label increased the model accuracy from 94.17% to 97.78%. Overall, this study reveals the temporal spectral response pattern of mild bruising in Xinjiang prunes, providing methodological references and data support for postharvest dynamic detection and sorting-strategy optimization.