Abstract: To address the issues of low detection performance, high missed detection rate, and false alarm rate in an infrared small-target detection caused by the small pixel size of weak targets and complex background interference, this study proposes an improved infrared dim and small target detection algorithm based on YOLOv12n, named CSM-YOLO. First, a CA-DEConv module was designed to replace the C3k2 module in the backbone network, enhancing the extraction capability of edge and contour information for weak and small targets. Second, the SCSA attention mechanism was introduced to enable the model to focus dynamically on key regions of small target features. Finally, a multiscale feature fusion module was designed to adaptively fuse features across different scales, reducing the loss of small-target information in deeper network layers and improving the overall detection performance. Experimental results on the public datasets SIRST, SIRST V2, and NUDT-SIRST demonstrate that the proposed CSM-YOLO improves mAP50 by 3%, 13.3%, and 13.2%, respectively, and mAP50:95 by 0.8%, 5.9%, and 15.5%, respectively, compared with the baseline YOLOv12n model, while also reducing the total number of parameters by 3.5%.