Abstract:The task of infrared small target detection is a significant subject in the field of military surveillance. Addressing the issue of computational redundancy in existing methods, this paper proposes an infrared weak and small target detection model based on an efficient layer aggregation network, achieving efficient and accurate infrared small target detection. First, the traditional residual structure in the classic U-Net architecture is replaced with an efficient layer aggregation network to enhance the feature extraction capability of the network while reducing computational redundancy. Then, features at different scales are fused to optimize gradient propagation in deep networks. Finally, a hybrid loss function is employed to provide diverse training signals, thereby accelerating model convergence. Experimental results demonstrate that the proposed infrared weak and small target detection method based on the efficient layer aggregation network can effectively detect infrared small targets in complex scenarios. On the NUDT-SIRST data set, the method achieves an IoU metric of 90.88% and a Pd metric of 99.36%, reaching the optimal level in comparison with mainstream models.