Research on Open Set Recognition Method for Aerial Infrared Targets

This paper proposes a classifier for open set recognition of aerial infrared flying targets based on sparse representation theory. Firstly, features of infrared images are extracted by using the pre-trained VGG19 neural network and HOG features, and then a fused over-complete dictionary is constructed. Subsequently, the sparse coding and reconstruction error of the training samples are calculated, and the reconstruction error vector is clustered. For unknown samples, their sparse coding coefficients and minimum reconstruction error are computed, and their categories are determined based on the clustering results. Experimental results demonstrate that this method can effectively distinguish between known and unknown categories, with the open set recognition performance outperforming the existing method.