An infrared image edge detection method based on edge characteristics combined with K-means is proposed in this study to solve the problem of edge extraction caused by the blurring of infrared image edges. First, human visual characteristics are combined with gray distribution characteristics at the edge of the infrared image to construct a data set reflecting its structural characteristics. Second, K-means is used to classify the data set into edge and non-edge points to extract the image edges. Third, the edge is refined using a two-step method to achieve infrared image edge detection. The experimental results show that the proposed method can extract the complete external contour of the infrared image through the adaptive threshold and retain the internal edge information, which can extract the weak edge and effectively suppress noise interference.