Computer vision has had a hard time for a long time finding small infrared targets in scenes with a lot going on. To improve the rate at which small infrared targets are found in these kinds of scenes, a new network called YOLO-IRIS has been suggested and improved based on YOLOv8. The Space-to-Depth Convolution (SPD-Conv) module is added to improve spatial precision, which makes it easier for the model to pull out target features at different sizes. The Efficient Multi-Scale Attention (EMA) system is built into the design. This improves the accuracy of feature selection and makes it possible to find small targets quickly. A better design for the sensing head makes it easier to do calculations and makes it more accurate when looking for small objects. An improved intersection over union (EIoU) loss function takes into account error in localizing small targets in complex backgrounds, which makes localization even more accurate. On a self-made dataset based on Anti-UAV410, YOLO-IRIS does very well. The average accuracy (mAP) is 8.04 percent higher than with the original YOLOv8s, and the number of parameters is 67.2 percent lower than with the original. When we look at these differences more closely, we can see that YOLO-IRIS has a higher average precision (mAP) than YOLOv5s and a lower average precision (mAP) than YOLOv6s. This is even though the number of factors has been cut down to 82.1% and 39.8% of their original amounts. These changes show that YOLO-IRIS works well and can be used to find small infrared targets in backgrounds with a lot of other things going on. This makes it an important answer in this field.