Abstract: Infrared target detection is one of the core technologies in infrared search and tracking systems. In complex backgrounds, infrared target signals are weak and there are numerous irregular sources of interference, which can easily lead to false alarms. To address this issue, this paper proposes an infrared small target detection algorithm that combines graph representation learning and multi-feature fusion. Initially, morphological methods are used to extract candidate target regions. Then, considering irregular false alarm sources and targets are difficult to represent visually in a coordinated manner, the candidate target regions are transform from the image domain to the graph domain to extract both handcrafted features based on images and deep features based on graph representation learning. Finally, a fully connected network is used for feature fusion and classification, thereby filtering out the false alarm regions and obtaining the target regions. The performance comparison experiments are conducted on a public infrared small target dataset, and the results show that the proposed algorithm has good detection performance in complex scenarios.