剖线强度梯度变化的作物软硬变化区精确划分

目的]软硬变化区的划分是进行遥感变化检测技术识别农作物的基础。为解决人工判读方法中存在人为主观因素影响以及自动判别和交互式判别方法对不同研究区的适用性问题,该文提出基于剖线梯度变化(Profile based Gradient Change Magnitude,PGCM)进行软硬变化区划分的方法。方法]该研究选择破碎农业景观种植区域为研究区,计算拔节期Quick Bird影像和播种期模拟影像两个时相的变化强度,从作物地块内部向外绘制剖线,利用剖线强度的梯度变化确定硬变化区(Hard change region,HCR)、软变化区(Soft change region,SCR)和未变化区(Non change region,NCR)3者间的划分阈值。结果]从识别结果来看,PGCM能够有效在地块边界处探测到软变化像元,进而确定HCR、SCR。像元分辨率在5~60 m不同尺度下,识别HCR区混入比例为11%~16%,混入比例随着分辨率下降而降低;NCR区混入比例为3%~4%,受分辨率尺度影响不大;SCR区识别比例为74%~86%,识别精度较高,识别结果与冬小麦空间分布结果保持一致。结论]PGCM方法能够自动、便捷地确定阈值,摆脱人工判定的主观性,有效地划分出HCR、SCR和NCR 3个区域,为进一步HCR、SCR区内的作物识别提供基础。

SOFT AND HARD CHANGE REGION IDENTIFICATION OF CROP BY PROFILE BASED GRADIENT CHANGE MAGNITUDE METHOD

Acquiring crop acreage timely and accurately is important to adjust planting structure and improve agricultural management. Remote sensing has played a significant role in crop acreage detection due to its large coverage and short revisit detection. Dynamic change detection with multi temporal images is much better than single image detection in terms of reducing the omission and commission. Currently, the integrated method of soft and hard change detection method is effective at mapping crop distribution. Extraction of soft change regions (SCR) and hard change regions (HCR) is the basis for change detection method to map crops by remote sensing. To avoid the subjectivity in threshold based method and the limitation of large area applicability over different regions in the interactive threshold decision methods, profile based gradient change magnitude method (PGCM) is developed to identify the HCR, SCR and no change regions (NCR). The threshold values of such three regions are determined based on the gradient change magnitudes, which is calculated along a profile from inside to outside of the target crop parcels. The study area is located at Daxing districts of Beijing with the area of 10 km×10 km. The fields in this area are fragmented and complicated with winter wheat, vegetable, and fruit tree staggered growing, which cause great difficulties for remote sensing identification. The QuickBird image with good quality, on 6 October 2005 (at sowing stage) and the simulated QB image (at jointing stage) were chosen. The simulation image was made according to QB image. The results verified that PGCM was capable of identifying the SCR which existed at the parcel boundary and HCR and SCR pixels according to threshold values. At different pixel resolution of 5 m to 60 m, HCR error varied from 11% to 16%, mix proportion of SCR was from 3% to 4%, with an accuracy of SCR 74% to 86%. The identification results were accordance with the actual spatial distribution of winter wheat. PGCM as an efficient model could determine threshold automatically and divide the whole region into HCR, SCR and NCR in avoid of the disadvantage of the subjective thresholds, which had potential to help crop identification in HCR and SCR further.