Yellow peaches are rich in nutrients and very popular with consumers. However, they are susceptible to bruising during processing and transportation, causing great economic losses to farmers and sellers. When using hyperspectral technology to detect fruit bruises, researchers only use spectral information or image features to build a bruise detection model but rarely use spectral fusion technology for fruit bruise detection. Moreover, spectral reflectance is easily affected by external stray light, some real information will be lost, and the image contains little information, It is difficult to identify images only relying on a few specific bands accurately. Therefore, to achieve an accurate classification of the degree of bruising of yellow peaches, different treatment methods can be developed according to the different degrees of bruising, and economic damage can be reduced. This study proposes a model that uses the spectral information of hyperspectral combined with image features to detect yellow peaches with different degrees of bruising. In this experiment, 180 yellow peaches were used as experimental samples. Firstly, the images of lightly, moderately and severely bruised yellow peaches were collected using hyperspectral images, A 100×100 pixel area was selected as the region of interest at the bruise location of each yellow peach, and the spectral information of the bruised area was extracted using ENVI4.5 software. Then, the principal component analysis (PCA) algorithm was used to reduce the dimensionality of the collected hyperspectral images, and the PC1 image was finally selected as the principal component image of the bruised yellow peaches among the first 5 PC images. The images corresponding to 6 characteristic wavelengths were selected as the feature images according to the weight coefficient curve of PC1 image, and the average grey value was used as the image feature of the bruised yellow peaches. Finally, PLS-DA models for yellow peach bruising were established using spectral information, image features and spectral information combined with image features respectively, and the performance of each model was evaluated using the correct classification rate. The results showed that the PLS-DA model based on spectral information combined with image features had the best discriminative effect, with classification accuracy of 85%, 90% and 100% for light bruises, moderate bruises and severe bruises respectively, and the acorrectoverall rate reached 91.7%. In order to further improve the accuracy and operational efficiency of the PLS-DA model, this paper uses the competitive adaptive reweighted sampling (CARS) algorithm to filter the spectral data in the fused data for the feature bands. The model built using feature bands combined with image features had the best classification prediction, with 95%, 90% and 95% correct predictions for lightly bruised, moderately bruised and heavily bruised yellow peaches, respectively, and acorrectoverall rate of 93.3%. In conclusion, this study shows that it is feasible to establish a PLS-DA model based on hyperspectral spectral data combined with image features to detect the degree of bruising of yellow peaches, which provides a theoretical basis for the post-harvest treatment of yellow peaches.