小波包-局部最相关算法提高土壤有机碳含量高光谱预测精度

发布时间：2018-02-26 08:44

本文关键词： 光谱分析土壤有机质小波包局部最相关　出处：《农业工程学报》2017年01期 　论文类型：期刊论文

【摘要】：高光谱遥感可以实现水稻土排水期有机碳含量的快速预测,但土壤反射率受多种噪声的影响,有机碳光谱信号探测受阻,预测模型性能低下,如何在去除噪声的同时最大限度地保持有机碳光谱信号十分重要。以原状新鲜水稻土为研究对象,采用Bior1.3小波系对反射光谱进行1~7层小波包变换,通过相关分析确定最大分解层;将原始反射率至最大分解层以内的各层光谱相关系数组成相关系数集,采用局部最相关算法(local correlation maximization,LCM)构造土壤有机碳最优光谱;最后基于最优光谱建立有机碳含量偏最小二乘预测模型并进行分析。结果显示:1)随着小波包分解层数的增加,土壤反射率与有机碳含量的相关性不断增强,到第6层达到最高,确定为小波包最大分解层;2)基于LCM构造的最优光谱比未去噪光谱平滑,比小波包去噪光谱保留了更多光谱细节;3)未去噪光谱、小波包去噪光谱和LCM最优光谱有机碳预测模型的验证决定系数分别为0.693、0.727和0.781,均方根误差为1.952、1.840和1.679 g/kg,残留预测偏差为1.85、1.97和2.17。小波包-局部最相关算法在去噪同时有效保持了土壤有机碳光谱信号,可提高水稻土有机碳含量高光谱预测精度。
[Abstract]:Hyperspectral remote sensing can quickly predict the organic carbon content in paddy soil during drainage period, but the soil reflectivity is affected by many kinds of noise, the detection of organic carbon spectrum signal is blocked, and the performance of prediction model is low. It is very important to keep the organic carbon spectral signal while removing noise. Taking the raw fresh paddy soil as the research object, the Bior1.3 wavelet system is used to transform the reflected spectrum into 1 / 7 layers of wavelet packet. The maximum decomposition layer is determined by correlation analysis, and the spectral correlation coefficients from the original reflectance to the maximum decomposition layer are formed into the correlation coefficient set, and the local correlation maximization algorithm is used to construct the optimal spectrum of soil organic carbon. Finally, the partial least square prediction model of organic carbon content based on the optimal spectrum is established and analyzed. The results show that the correlation between soil reflectivity and organic carbon content increases with the increase of wavelet packet decomposition layers, and reaches the highest in the sixth layer. The optimal spectrum based on LCM is smoother than the unnoised spectrum, and retains more spectral details than the wavelet packet denoising spectrum. The validation decision coefficients of wavelet packet denoising spectrum and LCM optimal spectral organic carbon prediction model are 0.693 ~ 0.727 and 0.781respectively, the root mean square error is 1.952 ~ 1.840 and 1.679 g / kg, and the residual prediction deviation is 1.85 ~ 1.97 and 2.17 respectively. The wavelet packet-local correlation algorithm has denoising at the same time. To maintain the soil organic carbon spectral signal, It can improve the precision of hyperspectral prediction of organic carbon content in paddy soil.
【作者单位】：南京农业大学资源与环境科学学院;
【基金】：中央高校基本科研业务费专项资金(KYZ201522) 江苏高校优势学科建设工程项目国家自然科学基金项目(41571171)
【分类号】：S153.6

【相似文献】