Aimed at the problem of unsupervised target detection in hyperspectral image, a target detection method using projection pursuit (PP) based on chaotic particle swarm optimization (PSO) is proposed. Chaotic PSO can speed up the process of PP and get more accurate optimal projection direction. Adaptive band selection is used for the dimensional reduction of hyperspectral image. Skewness and kurtosis which are susceptible to outliers are chosen to design the projection index. And chaotic PSO is applied to search for optimal projection direction. Thus the target information can be projected into low-dimensional space effectively. The target is extracted from projection images by histogram segmentation. Experiments with qualitative and quantitative evaluation are carried out for many images, and the detection results of the proposed method are compared with those of genetic algorithm PP method and RX method. The results show that the proposed method detects target in hyperspectral images more effectively and significantly reduces the running time.