基于高分1号影像的森林植被信息提取

实时最新森林植被信息的提取是林业航空植保作业的必要前提。论文以安徽省蚌埠市为研究区域,探讨了基于高分1号卫星遥感数据在亚热带农林植被混合地区的森林植被信息提取。根据植被物候信息差异选择了提取森林植被信息的5个关键时期高分影像,采用分区决策树方法监测森林植被的空间分布和面积信息,并与未分区决策树法的提取结果进行比较。结果表明：采用分区决策树法和未分区决策树法对于大中尺度森林植被信息提取的总体精度均优于85%。但分区决策树森林植被提取总体精度达到90.72%,较未分区决策树法提高3.80%、4.65%,Kappa系数达到0.81,较未分区决策树法提高约0.07~0.10,结合植被物候信息的分区决策树森林植被提取法好于未分区决策树法,能够满足林业航空植保作业的精度需求。具有较高空间分辨率、宽覆盖、短重访周期的高分1号影像,对于大区域的林业航空植保当年最新森林植被信息的提取表现出较大的潜力。

Extraction of Forest Vegetation Information Using GF-1 Imagery

Accurate and up-to-date forest vegetation mapping can provide a better understanding of forest resources and support decision-makers in implementing sustainable forest management. Unfortunately, the distribution information of forests plantation with high accuracy and fine spatial resolution is still not yet conveniently available. Remote-sensing technologies are common used in mapping forest vegetation owing to their real-time data acquisition ability. However, extraction of forest vegetation information using single date remote-sensing imagery has been unsuccessful since the existence of similarity in spectrum feature between forest and field crops. Combination of seasonal variations of spectral response and phenological differences between forest vegetation and field crops presents a unique opportunity for forest mapping. Therefore, a method for extracting forest vegetation using multi-temporal GF-1 imagery was proposed and validated in Bengbu City. Based on the phenological changes of forest and dominant field crops in the study area, the whole region was separated into 2 sub-regions (sub-region A and sub-region B), and 5 phases of GF-1 imagery were utilized. Then, 2 sets of decision rules were built and applied to the corresponding sub-regions. In addition, we implemented forest extraction by non-partitioned decision tree for comparative analysis. The results show that the overall accuracy of both partitioned and non-partitioned decision trees are over 85%, which means that decision tree method using multi-temporal GF-1 imagery can acquire good accuracy when extracting forest vegetation in the large scale and mesoscale. Partitioned decision tree achieves overall accuracy of 90.72% and kappa coefficient of 0.81, which are 3.80%-4.65% and 0.07-0.10 higher than the overall accuracy and kappa coefficient of non-partitioned decision tree, respectively. Free GF-1 imageries with a fine spatial resolution, wide coverage, and low revisit period have great potentials in forests extraction which can benefit forestry aviation plant protection.