Currently, research on unsupervised single-modality pedestrian re-identification mainly focuses on visible light images. With the proliferation of new infrared cameras, unsupervised infrared pedestrian re-identification also demonstrates its research value. Due to the low contrast and lack of color texture details in infrared images, global information is crucial for infrared pedestrian re-identification. This paper designs an unsupervised infrared pedestrian re-identification network based on F-ResGAM. The network first uses wavelet transform for image pre-processing to enhance feature extraction capabilities, and then introduces GAM (Global Attention Mechanism) in the ResNet50 network structure to focus on more global information. Furthermore, due to the high noise in infrared pseudo-labels, this paper proposes a group sampling strategy based on sample expansion (GSSE) to further optimize the generation of pseudo-labels, thereby improving the model's recognition accuracy. Experimental results show that the optimization methods proposed in this paper effectively enhance the accuracy of unsupervised infrared pedestrian re-identification, especially with a significant improvement in the rank metric.