To address the challenges of difficult feature extraction and low contrast in infrared small-target images, an improved attention mechanism is proposed. First, a parallel dual-channel reverse attention mechanism is designed. Based on the reverse attention concept, one branch processes features in the order of spatial attention followed by channel attention, while the other branch follows the reverse order—channel attention followed by spatial attention. These two independent branches are used together to merge their outputs. Second, a parallel dual-channel reverse attention mechanism is introduced into the Res2Net structure, and an improved region strength level module is added to Res2Net. Third, the loss function considers the global and local constraint loss functions. The simulation results show that improving the attention mechanism has good visual effects, and the accuracy and ROC detection performance are better than those of other algorithms.