Aiming at hyperspectral target detection in the unknown environments, an automatic target detection algorithm based on projection is presented in this paper. Through orthogonal projection, the approach suppresses part of interference information beforehand and finds the potential targets using unsupervised automatic target search method from the data processed before. After that, it projects the data into the subspace spanned by the potential targets to strengthen the target information and extracts the targets with matching method finally. The algorithm effectively lessens the bad effect of interference on target detection and reduces the range of target searching. The experimental data are processed using our method, and the satisfying results are achieved.