论文部分内容阅读
目前,高光谱数据精细分类面临两方面问题:一方面,传统单纯利用光谱信息的分类往往难以满足各应用行业对精度的需求,另一方面,基于像元的分类结果受制于椒盐噪声,影响其有效应用。为此,提出了一种基于植被特征库构建与优化的高光谱植被精细分类策略。首先,从高光谱影像中的原始光谱特征出发,结合灰度共生矩阵和局域指示空间分析两类纹理特征,并有针对性地加入了对植被叶绿素、胡萝卜素、花青素和氮素叶面积指数等理化参量敏感的光谱指数特征,构建了完备的植被特征库,以提高植被类别间的可分性;进而对植被特征库进行光谱维优化,提出了基于类对可分性的光谱维优化算法,选择对各类对具有最高识别能力的特征波段,通过迭代使各类别间均达到较高的区分度,并利用最优索引因子法进一步降低数据冗余,以提高分类效率;在进行植被特征库空间维优化时,主要基于地物分布通常具有一定的空间连续性这一理论,提出了基于邻域光谱角距离的植被特征库空间维优化算法,以去除分类结果中的椒盐噪声,提高分类精度和分类图像平滑度。基于航空高光谱数据的植被精细分类验证表明,该方法可以显著提高分类精度,在作物品种识别、精准农业等方面将具有广泛的应用前景。
Currently, the fine classification of hyperspectral data faces two problems: on the one hand, the traditional classification of spectral information is often difficult to meet the accuracy requirements of various application industries. On the other hand, pixel-based classification results are subject to salt and pepper noise, affecting its Effective application. Therefore, a new classification strategy of hyperspectral vegetation based on vegetation feature database was proposed. First of all, based on the original spectral features of hyperspectral images, two types of texture features were analyzed by combining gray level co-occurrence matrix and local indication spatial analysis, and the effects of vegetation chlorophyll, carotene, anthocyanins and nitrogen leaves Area index and other physical and chemical parameters sensitive spectral index characteristics, build a complete vegetation feature library to improve the separability between vegetation categories; and then spectral feature optimization of vegetation feature library, based on the class of separable spectral dimension Optimization algorithm, select the characteristic band with the highest recognition ability for all kinds of pairs, and achieve a higher degree of differentiation among all the categories through iteration, and use the optimal index factor method to further reduce the data redundancy so as to improve the classification efficiency; Based on the theory that the spatial distribution of vegetation usually has certain spatial continuity, this paper proposes a spatial dimension optimization algorithm based on the spectral spectral distance of the neighborhood to remove the salt and pepper noise in the classification results, Improve the classification accuracy and classification of image smoothness. The fine classification of vegetation based on hyperspectral data shows that this method can significantly improve the classification accuracy and has broad application prospects in crop variety identification and precision agriculture.