林业是生态建设的主体，在维护生态平衡中起决定性作用。近年来，随着遥感技术的发展和遥感技术在林业的深入应用，其成为国家获取林地资源信息的一种有效的手段。但相关研究表明，林地类型信息获取中仍存在精度欠佳、新方法应用少、自动化程度低、详细度和可信度差等技术难点。 本文主要探讨在大面积复杂林区，基于红边波段卫星数据，高空间分辨率遥感影像开展林地类型精细识别和变化监测技术，以促进具有红边波段卫星数据在林地类型调查与监测方面的深入应用，更为拥有自主产权且具有红边波段的GF-6卫星数据在林业中的应用打下坚实基础。具体研究内容与结果如下： （1）以内蒙古根河市根河生态站为重点研究区，基于2016年7月的RapidEye、GF-1 PMS和Landsat 8 OLI遥感影像，综合利用影像的光谱特征、纹理特征与根河森林资源小班数据以及2016年林地类型外业调查样本数据，分别对三种遥感数据源采用最大似然法（MLC)和支持向量机 (SVM）、基于EnMAP-Box的支持向量机（ImageSVM）和随机森林（ImageRF）分类方法、归一化红边指数（NDRE）参与下支持向量机分类和基于样本的面向对象分类法进行林地类型精细识别。最后以外业调查数据和根河森林资源小班数据作为检验样本对不同方法的林地类型分类结果进行精度验证，通过精度验证混淆矩阵进行分类结果评价。结果表明：1）ImageRF和ImageSVM分类法获得的林地类型分类信息精度最高。基于高分辨率的RapidEye和GF-1PMS遥感影像效果更明显，ImageRF和ImageSVM的总体分类进度相对于传统的SVM和MLC均提高了6%以上；基于中分辨率的OLI遥感影像，总体分类精度也均提高了1%左右；2）在林地类型精细识别中，携带红边波段信息的RapidEye影像比无红边波段信息的Landsat OLI和GF-1 PMS影像具有更高的识别精度和可分离性；3）在相同条件（影像类型相同、数据范围相同、运行环境相同）下，ImageRF与ImageSVM分类法精度基本一致，但ImageRF方法费时少、效率高； （2）针对携带红边波段信息的RapidEye遥感影像，包含归一化红边指数NDRE的SVM分类与传统的SVM分类的精度由84.08%增长到91.69%。由此可知林地类型信息对红边波段较为敏感，可极大的提高林地类型的识别精度；基于样本的面向对象分类法分类精度可满足林地类型精细分类要求，但是其自动化程度低、费时费力、受人为因素影响较大，不适合大区域林地类型精细识别研究。 （3）以内蒙古大兴安岭地区为整体研究区，基于2008~2015年夏季的Landsat TM/OLI遥感影像、辅助数据和外业实地样本点，开展EnMap-Box的随机森林分类后林地类型变化监测技术研究。结果表明：EnMap-Box支持下的随机森林分类后比较法精度较高。总体而言，研究期间内蒙古大兴安岭林地面积从2008年的73.19%增加到2016年的74.08%，生态状况不断改善。2008~2015年间，疏林地向阔叶林和针叶分别转换了2.13%和1.27%；1.01%和9.43%的草地转换为阔叶林和针叶林；有2.72%和2.11%的未利用地分别转换为阔叶林和针叶林。天然林资源保护工程的实施，是内蒙古大兴安岭林地面积增加的重要原因之一。 （4）以江西省吉水县为验证研究区，利用2008年Landsat TM和2016年GF-1多光谱遥感影像，结合2016年7月林地类型野外调查数据，对本文研究的变化监测技术的适用性进行验证。结果表明：2008和2016的总体分类精度分别为81.43%、83.89%，分类结果可靠性高；森林覆盖率由62.83%提高到63.86%，与实际森林增加量相符合，同时各地物类型所占百分比也均与吉水县实际相符合，故该方法在中国中部地区同样具有很好地适用性。

Forest is the main body of ecological construction, which mainly determined the ecological balance. In recent years, as the development of remote sensing, it is broadly used in forest, and it has become an accurate and high efficient method for monitoring the forest dynamics. However, the drawbacks of forest classification still exist due to the imperfect accuracy, low application frequency of new method, low automation and low reliability. In order to meet industry application needs of the national, this paper discussed the fine classification of forest land types based on high spatial resolution imagesin the largely complicated forest area, which locates in Genhe ecological station, Genhe City, Inner Mongolia Autonomous Region, with the support of the red-edge band satellite data to promote the deep application of the red-edge band satellite data in the investigation and detection of forest land type GF-6 satellite data with independent property rights and red side band lay a solid foundation in forestry applications. Firstly, the fine forest classification was conducted using Maximum Likelihood Classification (MLC), Support Vector Machine Classification (SVM), Image SVM based on IDL, and Random Forest (RF), and SVM with the participation of the normalized Red Edge Index, and object-oriented classification based on sample, based on multi-source data, including RapidEye, GF-1 PMS and Landsat 8 OLI remote sensing images obtained in July 2016; and other information spectral feature, texture and forest classification survey data. Finally, the results of field survey and the second forest resource survey data were used as the test samples to verify the classification results of different types of forest land types, and the classification results were evaluated by precision verification. The results showed that: ImageRF and ImageSVM have high precision for forest type information extraction. The high-resolution remote-sensing images of RapidEye and GF-1PMS are more effective. The overall classification accuracy of ImageRF and ImageSVM is more than 6% higher than that of traditional SVM and MLC. But, the medium-resolution remote sensing image, Landsat 8 OLI, the overall classification accuracy increased by about 1%. Meanwhile, in the fine identification of forest land type, the RapidEye image carrying red-edge band information has better recognition precision and separability than Landsat 8 OLI image with no red edge band information. Under the same conditions, including the same image type, the same data range and the same operating environment, ImageRF and ImageSVM classification accuracy is basically the same, but ImageRF method is less time-consuming and efficient. For the RapidEye remote sensing image with red band information, the precision of the SVM classification including the NDRE and the traditional SVM classification increased from 84.08% to 91.69%. It can be seen that the forest type information is more sensitive to the red band, which can greatly improve the recognition accuracy of forest types. The sample-based object-oriented classification method can meet the fine classification requirements of forest land types, but its low degree of automation, time-consuming and labor-intensive and subject to human factors, it is not suitable for for large-scale forest land type fine-grained research. Secondly, this study investigated the changes of forest land types using EnMAP-Box model with the random forest(RF) based on Landsat Thematic Mapper 5 (TM) and Landsat 8 OLI during 2008 and 2015 over Inner Mongolia Daxinganling area. The results showed that the classification based on EnMAP-Box model with the random forest(RF) was a more suitable method for the changing detection. Specially, the coverage of broad-leaved forest, coniferous forest and mixed forest increased, a slight decrease in sparse forests. The forest land increased from 9,619,700 hectares to 9,737,687 hectares, with an increase of 116,900 hectares. The cultivated land, water and construction land also increased, while the grassland, wetland and unutilized land decreased. Overall, during the period of study, the area of forest land in the Greater Xing'an Mountains in Inner Mongolia increased from 73.19%(2008) to 74.08%(2016), and the ecological situation continued to improve. From 2008 to 2015, the Sparse forest land converted 2.13% and 1.27% to broad-leaved forests and needles respectively; 1.01% and 9.43% of grasslands were converted to broad-leaved forests and coniferous forests; there were 2.72% and 2.11% of unused land were converted to broad-leaved forests and coniferous leaves, respectively. The implementation of the natural forest resources protection project is one of the important reasons for the increase in the forest area in the Greater Hinggan Mountains in Inner Mongolia. Finally, this study applied EnMAP-Box model with the random forest(RF) based on Landsat Thematic Mapper 5(TM) and Chinese domestic high resolution satellite data, Gaofen-1, to investigate the changes of land cover types between 2008 and 2016 at Jishui county of Jiangxi province, the Central China, that is a validation study area. The results showed that the overall classifications of 2008 and 2016 are 81.43% and 83.89%, respectively, and the classification results were reliable. The total forest coverage rate increased 1.03 per cent from 2008(62.83%) to 2016(63.86%), which was in accordance with the actual forest increase. At the same time, the percentage of various types of objects also conforms to the actual situation in Jishui County. Therefore, this method also has a good applicability in the central region of China.