Infrared dim target detection via hand-crafted features and deep information combination

Infrared target detection is the key technology of infrared search and tracking system. In the case of complex background, dim target detection face difficulties such as high algorithm complexity and low robustness. To solve these problems, an infrared dim target detection method via handcrafted features and deep information combination is proposed. Firstly, an adaptive Top-Hat segmentation is used to extract candidate target region, then 128-dimensional deep learning features and 106-dimensional hand-crafted features are calculated for candidate region respectively. Finally, features are concatenated and input into a 4-layer fully connected network for feature fusion and classification to exclude false alarm sources. Experiments show that the proposed method can detect infrared dim targets accurately and robustly in various scenes.