Hyperspectral image target detection via integrated background suppression with adaptive weight selection

Hyperspectral image (HSI) target detection is an important technology, but is faced with the critical challenge of complex background interference. The background clutter is difficult to effectively suppress, which often results in a relatively poor detection performance being obtained. To address this problem, a new target detection method via integrated background suppression with adaptive weight selection (TDIBS_AWS) is proposed in this paper. TDIBS_AWS has the following capabilities: (1) it explores the great difference between the target and background with two different background suppression strategies: principal component analysis (PCA) and spectral unmixing (SU); (2) it applies adaptive weight selection based on the particle swarm optimization (PSO) algorithm to optimize the weighting coefficient of the integrated background suppression model; and (3) it takes full advantage of support vector data description (SVDD) to improve the target detection performance for the rest of the information, after the background and noise have been removed. Experiments were undertaken using synthetic data and a real HSI, and it was found that TDIBS_AWS generally shows a better detection performance than the other state-of-the-art target detection methods.