森林资源规划设计调查，简称二类调查，是查清森林资源现状的重要手段，是科学培育森林、合理经营和管理森林资源的基础性工作，也是调查森林区域，制定林业规划设计、林业生产计划、林木资产化管理和检查、评价林业政策、方针、法规等的执行效果，实现林业产业健康、可持续发展的重要依据，其中基础内容之一就是小班区划，小班是林业资源调查体系的基本单位，为森林资源规划设计调查提供基础资料。 传统的小班区划方式为人工对坡勾绘，这种基于人工区划的方法费时费力，解译过程也会掺杂一些主观因素，对于林分类型的判断也是挑战，因此，发展基于遥感技术的半自动化的高效区划方法，以得到一致连续的区划结果是十分必要的。 文章结合小班区划的若干原则，以黑龙江伊春市带岭区为试验区，采用ALOS多光谱和全色高分辨率数据为数据源，（1）基于多尺度分割算法，研究了不同遥感信息参与对分割效果的影响，并且基于样地信息设计了评价指数，评价了分割结果，选择出合适的分割方式及分割尺度；（2）基于分割后对象的光谱、灰度共生矩阵特征及DEM信息，在特征提取与选择的基础上，采用计算机模式识别的CART及SVM分类器，实现了林分类型（针叶林、阔叶林、针叶混交林、针阔混交林）信息自动提取；（3）在分割分类的基础上进行了结果的合并和平滑处理工作，生成了试验区小班边界，采用信息统计的方式，比较了基于计算机的区划结果与人工区划结果在若干定量化指标方面的差异，证实了基于多尺度分割以及林分自动分类识别进行区划方法的可靠性。 研究结果表明：（1）由于高空间分辨率影像可利用的光谱信息有限，传统的基于像元的分类方式很难满足林区林分类型识别的要求，在识别精度和精细程度上都低于面向对象的分类方法；（2）单纯依靠光谱信息的分割分类对林分类型的识别的能力有限，采用地面调查的验证数据结合PODiff指数进行评价得出，其分割对象的信息与样地的信息差异比较大，分割效果较差，而加入坡度、坡向信息的分割结果则能更好地与样地契合；对于分割尺度的选择，结合最优分割尺度的定义，利用SEI指数进行评价，比较研究了五种不同尺度在四种林分区域的分割效果，筛选出50为最优分割尺度；（3）开展了基于多特征的森林类型识别研究，分别比较了在仅有光谱信息、叠加纹理特征以及加入地形因子三种情况下的分类精度，结果表明，纹理信息对于不同林分类型的识别有很大作用，而地形因子对于特定类别有识别效果，不同尺度下获得的特征叠加分类结果印证了50的尺度下各林分识别精度达到最优，同时证明了SEI指数对于林分类型的可分离性是有指示作用的；（4）采用定量化评价的方式，从面积、周长等方面与人工区划结果进行对比，其结果的一致性程度达到80%以上，各林分面积吻合性平均为73.4%，这说明基于多尺度分割分类方式的计算机区划方法取代人工区划的可行性较高。

Forest resource inventory known as the secondary inventory is fundamental but significant in thoroughly investigating real-time situation of forest. And it is the base of silviculture, reasonable management and operation. It helps in establishing forest planning, designing, forest production plan, and forest capitalization administrative as well as evaluating the implements effect of policies, guidelines and laws. Depending on this work, the forest administrative could be healthy and sustainable. In the whole inventory process, the most elementary but important work is forest delineation which divided the forest area into small parts ---the unit of inventory system and is also the fundamental data for forest planning and designing. Traditional delineation was mostly based on manual force which was time and labor consuming, and also there were some interpretation mistakes lying in the result because subjective opinions would affect the result. Therefore, it is necessary to develop a semi-auto method depending on remote sensing technology to get a more accurate and consistency result. Considering the principles in delineation work, taking ALOS multi-spectral and panchromatic image of Dialing, Heilongjiang province as data source, we had some test on the following sides. (1) We researched on differences of multi information integrated in segmentation process and developed two indexes to evaluate the segment results according to which we chose the best segment parameters consisted of segment scheme and scale. (2) We identified four basic forest types including coniferous forest, broad-leave forest, coniferous mixed forest and mixed broad-conifer forest by SVM and CART classifier based on spectral, texture and DEM information extracted from segment results. (3) After some manual edit work like small object elimination and border smoothing on result produced by step (1) and (2), we got the final computer delineation. Compared with manual result in some quantitative indexes, we verified the delineation accuracy of the method we took in this experiment. The result showed that: (1) object-based classification applied in forest identification based on high spatial resolution image has more advantages than traditional pixel-based one because the more detailed texture features lies in the objects. (2) It is limited when segmentation only with spectral information. Through evaluating by true sample data, we found it was such different and the segmentation result was not that good. But when considering terrain information into segmentation, we got a closer result. According to the definition of optimized scale, we compared five different scales in segmentation and got five SEIs in stand for homogeneity and heterogeneity of each object. We found 50 as the best scale in forest delineation. (3) In the study on multi-feature forest identification, we had test on three individual situations which were spectral only, spectral and texture features and additional terrain information. The result shows the classification accuracy was higher when coming into the second situation and terrain information was really help in specific types. What’s more, we compared classification accuracy in different scales and verified 50 was the best scale and SEI index was effective to express the classification possibility. (4) Taking area, perimeter and other indexed as objective evaluation indexes, computer-assistant delineation was 80% close to manual one and the average area coherency of forest types was 73.4% which was such close to reference in spatial distribution. To make a conclusion, we thought it was so possible that this cost-saving and efficient method would replace the costive and subjective traditional one.