Existing small object detection algorithms have shown good detection performance. However, small object detection still faces significant challenges when it comes to decoupling foreground and background with similar colors and shapes. Therefore, this research presents an interactive cross attention augmented small object identification technique for tiny object recognition. Regions of interest (ROI) are extracted through a refined classification residual network. To solve the problem of foreground-background imbalance, the feature maps of the regions of interest recovered by the improved classification residual network are refined using the interactive cross attention UNet detection network. The Interactive cross attention UNet detection network effectively extracts edge transition information of both foreground and background, improving the detection accuracy. The conversion connection module connects the refined classification residual network and the interactive cross attention UNet detection network, enabling efficient feature transformation and collaboration between them. By utilizing the refined classification residual network and interactive cross attention augmented, semantic information, structural information, and global contextual information in the mouse images are effectively captured. Experimental comparisons are conducted on a collected dataset of 7133 mouse images. Under the same experimental settings, our algorithm outperforms three existing small object detection algorithms, exhibiting significantly better detection performance. Intersection over Union (IOU) outperforms by 1%, while the mean Average Precision (mAP) improves by 0.5%.