Target detection is an important part of infrared image processing, and the subsequent processing is directly affected by the detection results. On the basis of analyzing the characteristics of infrared images, an improved Top-Hat operator is adopted to suppress the noise points in infrared images. And according to traditional K-means clustering idea, K-means clustering target detection algorithm based on two-dimensional gradient information is proposed. Experimental results show that the method has distinctly noise suppression effect, and the target in infrared images is well detected, which lays a good foundation for the subsequent image processing.