Abstract The instant and accurate burn assessment is very important for burn wound treatment. But due to the limited burn wound samples, the accurate classification of burn wounds is difficult. In this work, a full-field burn wound classification system is proposed using the near-infrared hyperspectral imaging (NIHSI) with the deep transfer features. The convolutional neural network (CNN) is applied to establish a nonlinear relationship between the hyperspectral image data and the burn severity. And the transfer learning strategy is used to boost the high classification accuracy. The burn wounds are created on bilateral sides of the abdomen of a bama miniature pig. The burn depth model (BdasNet) is pre-trained based on side A, and then fine-tuned with data of side B. The proposed model BdasNet can get the classification accuracy more than 96% for burn depth, which demonstrated the proposed method can perform an accurate classification of burn wounds and provide more practical references for clinicians.