Based on analyzing the infrared characteristics and shape characteristics of condensing tower that has special construction rules, an infrared target detection algorithm based on knowledge model is proposed. Firstly, based on the infrared characteristics of condensing tower, the intensity, orientation and local entropy features are extracted to construct the visual attention model, which is used to extract salient regions as regions of interest in the infrared image. Secondly, based on the shape characteristics of condensing tower, hyperbola shape model is constructed for condensing tower, structure feature edges are extracted in the salient regions and used to fit the hyperbola shape model, and relevant decision rules are constructed to confirm the targets. The recall and precision of the experiment on a set of air-born infrared images can reach up to 98.67% and 93.67% respectively, which demonstrates the excellent performance of the proposed algorithm. Moreover, since the reference image is unnecessary in the proposed algorithm, the requirements for the data preparation is reduced greatly, which improve the practicality of the algorithm.