Research On Target Recognition In Hyperspectral Imagery Based On Spectral Information

Hyperspectral image that is a new type of remote sensing data has become one of recent research focuses in this field, and used in military and civilian aspects widely. Hyperspectral image provides numerous bands, high spectral resolution and sufficient spectral information, which allow targets to achieve effective classification and identification possible. Therefore, research on how to fully tap the spectral information in the hyperspectral image and identify target exactly has important theoretical and practical value.Compared to conventional optical image and multispectral image, hyperspectral image has the advantage of the rich spectral information except spatial information, this paper introduces hyperspectral imaging mechanism first, and then takes two common materials: mineral and vegetation for example, discussing their specific application by hyperspectral image; the higher spectral resolution enable them to maintain some of the key features. Since hyperspectral image has emerged in the 1980s, regarding each application many information processing approaches have been proposed, hence how to assess these approaches effectively is a problem which must be resolved. This paper brings up the concept"hyperspectral image utility", and discusses several common performance evaluation algorithms systematically for target detection and recognition. This paper uses spectral information to identify the target from two aspects: spectral waveform matching and spectral characteristics. In research of target recognition algorithm in hyperspectral image based on spectral waveform matching, we first discuss several typical algorithms, and then analyzing the indication of spectral angle mapping, a combination spectral operator is constructed. The operator overcomes the disadvantages of traditional spectral angle mapping and spectral information divergence, takes use of their information complementary and creates their advantages. Finally, we compare experiments by typical algorithm and improved algorithm using AVIRIS San Diego Airport hyperspectral image and assess them by ROC curves and SUM standard. Experiments show that the improved algorithm can improve the recognition effect to some extent.In research of target recognition algorithm in hyperspectral image based on spectral features, we introduce the concept of spectral characteristics and several common absorption spectra characteristics, using which distinguish the refined targets in hyperspectral image better. Meanwhile, this paper form broad spectral characteristics, proposes a fusion recognition method based on spectral reflectance and derivative information in hyperspectral image. Derivatives are relatively insensitive to variations in illumination. The method proposed has a good performance for refined target recognition. Finally, upon the theoretical before, In view of the characteristic of hyperspectral image processing system such as high data throughput, processing speed and large memory capacity, this paper designed an embedded hyperspectral image processing system based on DSP&FPGA, which adopted USB interface as transmission, and selected TI's TMS320C6000 DSP and Altera's Cyclone II series FPGA. Firstly, the general block diagram of hardware implementation is given; secondly, the chips used are introduced; the circuit connects of FPGA and the working principles of system are described in detail. Finally, the system is verified by taking advantage of hyperspectral image target recognition algorithm.