Due to the long distance and complex background， it is hard for the infrared detecting and tracking system to find and locate the dim-small targets in time. The proposed method， ACE-STDN， aims to tackle this difficult task and improve the detection accuracy. First of all， an adaptive contrast enhancement subnetwork preprocesses the input infrared image， which is conducive for the low-contrast dim targets. Next， a detection subnetwork with a hybrid backbone takes advantage of both convolution and self-attention mechanisms. Besides， the regression loss is designed based on 2D Gaussian distribution representation instead of Intersection over Union measurement. To verify the effectiveness and efficiency of our method， we conduct extensive experiments on two public infrared small target datasets. The experimental results show that the model trained by our method has a significant improvement in detection accuracy compared with other traditional and data-based algorithms， with the average precision reaching 93.76%. In addition， ACE-STDN achieves outstanding detection performance in a multiclass object dataset and a general small object dataset， verifying the effectiveness and robustness.