基于高分辨率遥感影像的大麻作物信息提取方法研究

发布时间：2018-03-25 02:24

本文选题：决策树　切入点：面向对象　出处：《中国地质大学(北京)》2017年硕士论文

【摘要】：大麻不仅仅是一种工业原料还是毒品的主要原植物,其花叶中含有具有致幻成瘾作用的的四氢大麻酚(THC),容易被用来吸食。随着种植规模的扩大,我国大麻产量已居世界第一位,受我国的种植模式的影响,使得大麻种植分布零散,因此造成国家缺乏有效、准确的手段来收集大麻种植范围和面积信息。由于大麻品种繁多,种植条件各异,在我们前期毒品原植物调查与监测的实际工作中发现,大麻在种植过程中,容易发生生物学混杂等而导致含毒量变化,使得部分地方种植的大麻毒性成分(THC)含量超标,成为事实上的毒品原植物,为毒品泛滥留下了严重隐患。本文以内蒙古自治区达拉特地区为研究区,基于国产高分辨率的高分二号卫星遥感影像,利用两种不同的方法对大麻地块进行了信息提取。首先通过对研究区影像光谱特征、空间特征、语义特征等进行分析,选取相应的特征值作为决策树节点,构建大麻地块信息提取的决策树,应用ENVI软件平台对原始数据进行由大到小,由粗到细的分类过程,实现大麻地块的快速识别提取最终得到预想的分类结果。其次利用面向对象分类方法提取大麻地块信息过程中,以四叉树分割作为多尺度分割的预分割,以提高分割效率,尝试将归一化植被指数NDVI这一语义信息作为单独一个波段组合到影像中参与影像分割,最后基于单波段最优尺度计算模型的基础上,将全部波段共同参与模型计算,对最优尺度选择模型进行了改进。研究表明:1、面向对象的信息提取方法提取的结果相较于基于像元的决策树分类器提取的分类结果无论在用户精度、生产者精度还是总体精度上都有很大的提高。2、改进后的分割方法使得分割效率得到了很大的提升,使得大面积的大麻种植监测成为可能。3、将NDVI作为单独的一个波段参与影像分割也使分割精度得到了提高,使得植被与非植被的区分更为明显。4、改进后的最优尺度选择模型综合考虑了所有波段的特征,依据该模型得到的最优尺度更具科学性,从根本上保证了后续的信息提取精度。
[Abstract]:Cannabis is not only an industrial raw material but also the main original plant of drugs. Its mosaic contains the addictive tetrahydrocannabinol (THCC), which is easily used for smoking. With the expansion of cultivation scale, our country's cannabis production has ranked first in the world. Affected by the pattern of cultivation in our country, cannabis cultivation is scattered and, as a result, the country lacks effective and accurate means to collect information on the extent and area of cannabis cultivation. In our earlier investigation and monitoring of the original drug plants, we found that in the process of cannabis cultivation, it is easy to produce biological hybrids, which leads to changes in toxic content, which results in the content of toxic components of cannabis grown in some places exceeding the standard. This paper takes the Dharat region of Inner Mongolia Autonomous region as the research area, and bases on the high-resolution high-resolution satellite remote sensing image of high-grade II made in China. In this paper, two different methods are used to extract information from hemp plots. Firstly, the spectral, spatial and semantic features of the image are analyzed, and the corresponding eigenvalues are selected as decision tree nodes. The decision tree of hemp plot information extraction is constructed, and the process of classifying raw data from large to small, from coarse to fine using ENVI software platform is carried out. In order to improve the efficiency of segmentation, the presegmentation of hemp plots is achieved by using quadtree segmentation as multi-scale segmentation in the process of extracting information from cannabis plots by using object-oriented classification method in order to achieve the rapid identification and extraction of cannabis plots. The semantic information of normalized vegetation index (NDVI) is used as a single band to participate in image segmentation. Finally, based on the optimal scale calculation model of single band, all bands are involved in the model calculation. The optimal scale selection model is improved. The research shows that the result of the object oriented information extraction method is better than that of the decision tree classifier based on pixel, regardless of the user's accuracy. The producer precision or the overall precision have been greatly improved. 2. The improved segmentation method has greatly improved the efficiency of segmentation. This makes it possible to monitor cannabis cultivation on a large scale. 3. Using NDVI as a single band to participate in image segmentation also improves the segmentation accuracy. It makes the distinction between vegetation and non-vegetation more obvious. The improved optimal scale selection model takes into account the characteristics of all bands, and the optimal scale based on the model is more scientific, which basically ensures the accuracy of information extraction.
【学位授予单位】：中国地质大学(北京)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S563.3;S127

【相似文献】