基于空间特征的光谱分类算法研究

随着成像光谱仪器的广泛应用,利用数据立方体进行物质分类与识别成为一项重要的研究内容,分类算法对最终的目标识别准确度与精度具有很大的决定作用。目前常见的分类算法主要利用了光谱维信息,从光谱匹配的角度进行物质分类。由于仪器探测的物质反射光谱不仅反映了物质种类,还与物质表面的几何结构,表面粗糙度等有关,因此仅仅利用物质的反射光谱进行物质分类识别具有一定的误差。该文在利用可见光反射光谱进行分类的基础上,结合图像空间特征,对分类过程进行控制,达到提高分类准确度的目的。利用该分类算法进行真假叶片识别,结果表明其具有较好的空间连续性,很大程度上克服了麻点效应,验证了算法的有效性。

Research on spectral classification algorithm based on spatial feature

With the wide use of imaging spectroscopy, applying data cubes to classification and identification of materials has been developed to be an important research content. The classification algorithms play a vital role in accuracy and precision of object identification. The most common classification algorithms mainly make use of the information gained from spectral dimension and classify the materials based on spectral match. The material reflectance spectra collected by imaging spectroscopy is determined not only by the sorts, but also by the geometry structure and roughness of material surface, and so on. Then classification and identification algorithms only using the reflection spectra have errors to some extent. This paper puts forward an algorithm based on the common classification algorithms that controls the classification process by using the spatial feature of image to promote the correctness of classification. This algorithm was applied to identify the true leaves from the fake ones. The result shows preferable spatial continuity. To a great extent, the algorithm overcomes ma pixel domino effect, and is proved valid.