Abstract: In military applications of infrared imaging guidance, the ability to segment targets quickly and accurately in infrared images is of great significance for improving guidance precision. However, a deep learning-based real-time semantic segmentation algorithm can easily lose key information about the target during downsampling. To address this problem, this study developed a real-time semantic segmentation algorithm for infrared images based on a wavelet pooling tri-branch network. First, a Haar wavelet pooling module was designed to reduce the feature-map resolution and information loss during the downsampling process. This module first used a 2D Haar wavelet transform to decompose the input feature map into one low-frequency component and three high-frequency components (horizontal, vertical, and diagonal). These four components were then stacked along the channel dimensions to form a new feature map. Subsequently, the wavelet-domain features were enhanced through operations such as convolution, batch normalization, and activation. Second, the Haar wavelet pooling module was incorporated into a tribranch network designed to extract spatial, semantic, and boundary information. Thus, a real-time semantic segmentation algorithm for infrared images was constructed based on a wavelet-pooling tri-branch network. In an experiment using a self-made infrared interference aircraft dataset, the MIoU and MPA accuracy metrics had values of 90.49% and 94.47%, respectively, which were superior to those of real-time semantic segmentation algorithms such as PIDNet. The proposed algorithm reduced the parameter count and computational cost by 7.6% and 4.4%, respectively, compared to PIDNet, making it capable of meeting the real-time infrared-image-segmentation requirements.