中文摘要：

针对PSI（Pixel Shape Index）算法存在的弊端,本文对其加以改进,提出一种光谱阈值自适应的改进型像元形状指数（Improved Pixel Shape Index of Adaptive Spectral Threshold,IPSIAST）算法。相比原算法,新算法的改进主要体现在：1）在方向线生成阶段,为了充分考虑不同波段光谱特征的同质性存在的差异,每个波段数据层方向线的生成分开独立进行,提高了方向线生成的合理性;2）将每一条方向线长度定义为各波段数据层在该方向的方向线长度的加权和,以体现不同波段光谱特征之间同质性上存在的差异,从而提高像元PSI值的精确性;3）光谱阈值T1能根据局部区域内光谱特征同质性的大小而自适应地调整,降低了地物边缘处的对比度对像元形状指数计算的影响。最后,结合目视对比和5个定量评价指标,通过影像分割对比实验证明：相比PSI指数,IPSIAST指数能进一步提高影像的分割精度。

英文摘要：

Aiming at the defects of the PSI （Pixel Shape Index） algorithm, this algorithm has been modified accordingly , and an improved PSI algorithm of adaptive spectral threshold （ Improved Pixel Shape Index of Adaptive Spectral Threshold , IPSIAST） is proposed in this paper .The main advantages of the new algorithm are following：1） In order to enhance the rationality of direction line generation , the differences existed in the spectral feature homogeneity between the bands have been considered adequately . The direction line on each band layer is extended independently .2） Length of each direction line is defined as the weighted sum of the lengths about all direction lines along the same orientation on all band layers .The differences existed in the spectral feature homogeneity between the bands are further reflected by this way , and the accuracy of PSI about each pixel is improved .3） Spec-tral threshold T1 can be adjusted adaptively according to the spectral homogeneity of the local region , and the influence of the con-trast on the edge of the surface feature will be reduced during the computing process of the PSI. Finally, by using visual compari-son analysis and five quantitatively evaluating indexes , the results of experiments showed that compared with the PSI , the imagery segmentation accuracy was further improved through associating with the IPSIAST .