To study the impact of short infrared wave bands (1 100-2 500 nm) setting on crop classification accuracy, and to provide data and application support for designed new types of sensors, in the study, we took north part of Bei’an city, Heilongjiang province as the study area, and studied the impact of the short infrared wave bands of 1 560-1 660 nm (SWIR 1) and 2 100-2 300 nm (SWIR 1) on identification accuracy of two crops of corn (Zea may L) and soybean (Glycine max) by selecting OLI (Operational Land Imager) data carried by US LandSat-8. Based on the mono temporal and multi-temporal conditions, and on the basis of five wave bands of coastal blue (433-453 nm), blue (450-515 nm), green (525-600 nm), red (630-680 nm), and near-infrared (845-885 nm), with five wave band combinations without short infrared wave band were used. One additional short infrared wave band (SWIR 1) was involved, and two additional short infrared wave band (SWIR 1and SWIR 2) were also used, with a total of 6 alternative schemes. In the study, we analyzed the remote sensing identification ability of short infrared on two crops: corns and soybeans, based on the maximum likelihood supervised classification method. Mono temporal OLI image Data of August 7, 2014 were taken for the study and five multi temporal image data of June 13, June 29, July 15, August 7, and September 17 of 2014 were also taken. The 5 km × 5 km grids of the covered study area were taken as the study units and under the principle of equal probability of corn and soybean area ratio, and 21 ground samples were chosen as training samples. used A visual interpretation method was used for identification of crop within ground samples. Accuracy verification data were from the background investigation results of crop areas in the study area. These data were obtained by using visual correction based on Rapideye image automatic classification with a spatial resolution of 5 m. The result showed that under the condition of mono temporal image classification, introduction of short infrared wave band can greatly improve the classification capacity of corn and soybean. Under the condition of introducing one short infrared wave band, the overall classification accuracy has been improved from 87.0% to 90.8%, up by 3.8%, Kappa coefficient was improved from 0.74 to 0.82, with a significant decrease of “Pepper salt problem”, The user's accuracy of corn classification has been improved from original 85.4% to 91.5%, up by 6.1%. Mapping accuracy was improved from 89.6% to 90.3%. The mapping accuracy of soybean has been improved from original 84.5% to 91.5%, up by 7.0%, with user's accuracy improved from 88.9% to 90.2%. The separation degree of corn and soybean has been improved from 1.53 to 1.93, indicating that the short wave infrared wave band can remarkably improve the separation capacity of corn and soybean. Under the condition of multi-temporal image classification, the improvement of identification capacity of corn and soybean by introducing of short infrared wave band was limited. Under the condition of introducing one short infrared wave band, the overall classification accuracy was improved from 92.4% to 92.9%, only up by 0.5%, indicating that multi-temporal data, to some extent, can replace short infrared wave band on improving crop identification effect. Regardless of mono temporal or multi-temporal conditions, introduction of two short infrared wave bands did not show any significant changes on overall identification capacity compared with one short infrared wave band. Both correlations of five images on two short infrared wave bands exceeded 0.96, indicating that introduction of redundant bands with strong correlation has limited effect in improving crop area identification accuracy. The results have further quantified the separating capacity of short infrared wave bands on two crops of corns and soybean, and provided basis for wave band setting of domestic satellite short-wave sensors.%为研究短波红外波段（1100～2500nm）设置对于作物分类精度提高的影响，同时也为新型传感器设计提供数据支持和应用支撑，该文以黑龙江省北安市北部为研究区域，选择美国LandSat-8携带的陆地成像仪（operational land imager，OLI）数据，基于单时相和多时相条件下，在海岸蓝（433～453 nm）、蓝（450～515 nm）、绿（525～600 nm）、红（630～680 nm）、近红外（845～885 nm）5个波段的基础上，采用陆续增加1560～1660 nm（SWIR 1）、2100～2300 nm（SWIR 2）两个短波红外参与分类的方式，基于最大似然分类方法，比较分析了短波红外对玉米和大豆两种作物的遥感识别能力。结果表明，在单时相影像分类条件下，短波红外波段的引入可以在很大程度上提高玉米和大豆的分类识别能力，相比无短波红外参与分类，引入1个短波红外波段后总体分类精度从87.0%提高到90.8%，提高了3.8个百分点，Kappa系数由0.74提高到0.82，且“椒盐现象”显著减少。玉米分类的用户精度从原来的85.4%提升到91.5%，提高了6.1个百分点；制图精度从89.6%提升到90.3%。大豆分类的用户精度从88.9%提升到90.2%；制图精度从84.5%提高到了91.5%，提高了7个百分点。从分离度结果分析，玉米和大豆分类的分离度从1.53提高到了1.93，表明短波红外波段可以显著提升玉米和大豆的分离能力。在多时相影像分类条件下，短波红外波段的引入对于提高玉米和大豆的识别能力提高有限，引入1个短波红外波段条件下，相比无短波红外参与分类，总体分类精度从92.4%提高到92.9%，仅提高了0.5个百分点，表明短波红外波段并未提升多时相作物分类有效信息。从短波红外个数分析，无论在单时相还是多时相条件下，引入2个短波红外波段与1个短波红外波段面积提取总体精度没有明显变化，5景影像2个短波红外波段相关性都在0.96以上，表明相关性很强的冗余波段的引入，对农作物面积识别精度的提高能力有限。研究结果定量阐明了短波红外谱段对玉米和大豆两种作物的区分能力，为中国国产卫星短波传感器的波段设置提供了依据。
展开▼
