This paper evaluates the performance of YOLOv5 in detecting small targets and proposes structural improvements to enhance its capabilities. Firstly, an additional small targets detection head is integrated into the original three-head architecture. To enrich feature map channel information, the convolutional block attention module (CBAM) is embedded in the backbone, while the involution attention mechanism is incorporated into the neck. Furthermore, the original intersection over union (IoU) loss is combined with the normalized wasserstein distance (NWD) loss to reduce sensitivity to positional deviations in small targets. These modifications improve the ability of the model together to detect and localize small targets effectively.