阈值选择方法在高光谱图像目标检测中的应用

Application of Threshold Selection Method in Target Detection of Hyperspectral Image

根据约束能量最小化、自适应一致估计、光谱角匹配等理论,提出了利用传统图像分割中的阈值选取方法对高光谱图像进行目标检测的方法。通过不同的目标检测方法将高光谱图像转化为灰度值图像,利用直方图极点法、迭代阈值选取法、最大类间方差阈值法分别提取了各个方法结果灰度值图像的阈值,并根据阈值进行灰度值图像的分割,识别出高光谱图像中的目标。结果表明:3种阈值选取的方法都能较为准确地提取到目标的位置与形状大小信息,与传统方法比较,利用阈值提取的精度更高,误分率更低,其中自适应一致估计算法与最大类间方差阈值法配合结果的正确率最高。

The threshold selection in traditional image segmentation was proposed in the target detection process of hyperspectral remote sensing image.Method for target detection of hyperspectral images was proposed.The hyperspectral image was converted into gray value image by different target detection methods,and the threshold value of each method result gray value image was extracted by histogram pole method,iterative threshold selection method and maximum inter-class variance threshold method,according to the threshold value.The segmentation of the grayscale image was performed to identify the target in the hyperspectral image.The results show that the three threshold selection methods can extract the position and shape size information of the target more accurately.Compared with the traditional method,the threshold extraction is more accurate and the misclassification rate is lower.Among them,the adaptive consensus estimation algorithm and the result with the maximum inter-class variance threshold method has the highest correct rate.