بررسي تأثير روش قطعه‌بندي بر استخراج شبكه راههاي مناطق شهري در تصاوير ماهواره‌اي با قدرت تفكيك بالا

Analyzing the Effect of Segmentation Method on Road Network Extraction in Urban Areas from HR Satellite Imageries

استخراج راه از تصاویر سنجش از دور، یك روش سریع و ارزان برای بدست آوردن اطلاعات حمل­ونقل و بروزرسانی سیستم اطلاعات مكانی  است. تغییرات مداوم و سریع محیط­های شهری، نیاز به بروزرسانی منظم و یا بازبینی لایه­های شبكه راه در سیستم­های اطلاعات مكانی را افزایش داده است. در این تحقیق علاوه بر پیاده­سازی یك چارچوب كلی استخراج راه مبتنی بر شیء، تاثیر كیفیت روش قطعه‍بندی بر نتیجه نهایی سیستم استخراج راهها مورد بررسی قرار گرفته است. برای ارزیابی نتایج از داده­های واقعی سنجنده Worldview 2 مربوط به منطقه شوشتر در استان خوزستان استفاده گردیده است. نتایج بدست آمده كیفیت مناسب­تر قطعه­بندی به روش مالتی رزولوشن را گزارش نموده است. در ادامه با انجام عملیات هرس كردن نتایج بدست آمده بیش از 20% بهبود یافته است. نتایج نهایی پیاده­سازی قطعه­بندی مالتی رزولوشن و هرس كردن بر روی داده مورد استفاده، مقدار پارامتر صحت 88% و پارامتر دقت 85% را نشان داده است. همچنین انتخاب پارامترهای روش قطعه­بندی مالتی رزولوشن تحلیل شده است و تاثیر انتخاب صحیح این پارامترها بر نتیجه بدست آمده مورد بررسی قرار گرفته است.

Road extraction from remotely sensed imageries is a rapid and cost effective method for acquiring transportation information and updating GIS (Geographic Information System) systems. Fast and continuous changes of the urban environment, increase the necessity of regular updating or revising road network layers in GIS systems. The difficulties in the design of an automated road network extraction system using remotely-sensed imagery lie in the fact that the image characteristics of road feature vary according to sensor type, spectral and spatial resolution, ground characteristics, etc. Even for an image taken over a particular urban area, different parts of the road network reveal different characteristics. In the real world, a road network is too complex to be modeled using a mathematical formulation or an abstract structural model. The existence of other objects (e.g., buildings and trees) casts shadows to occlude road features, thus complicates the extraction process. In this research, a general object –based framework for road extraction is implemented, moreover the effect of selection of segmentation method on road extraction is analyzed. Image segmentation is considered as the first and crucial step of objects based image analysis, which aims to obtain the so-called homogeneous segments for succeeding feature extraction, classification, and higher level image analysis. Extensive research has been conducted in the area of image segmentation. Major categories of current state-of-the-art RS image segmentation methods can be classified as follows: 1) point/pixel based; 2) feature based; 3)edge based; 4) region based; 5) texture based; 5) hybrid and so on. Preprocessing as the first step in the proposed method is designed to improve the quality of the image and identify relevant image pixels for further processing. Then, object-based segmentation method is used to extract the initial road segments. Segmented objects are classified into a binary image which represents road and non-road classes. In the next step, skeleton of road objects are extracted. After Skeletonization, a compact approximation of line segments and curves in a vector format are implemented in vectorization step. Small branches in road network, which are produced hitherto and are not known as road, are removed in pruning step; and finally the proposed method is evaluated by comparing with reference road network (as ground truth), which are generated from the road vector data from the GIS or manually extracted road network. For evaluation of the proposed method, real data of Worldview2 sensor in Shushtar area in Khuzestan province-Iran is utilized. Three different segmentation method implemented in eCognition software are tested. In this research, 2 popular quality metrics defined in literatures will be adopted. These metrics include completeness and correctness. Better quality using multi-resolution segmentation method is achieved. Pruning extracted road network leads in above 20% improvement in results. Final results - after multi-resolution segmentation and pruning- show 88% correctness and 85% completeness as evaluation criteria. In addition, selection of multi-resolution segmentation parameters is appraised and the effects of these parameters are assessed. This paper generally emphasizes on the role of image segmentation quality on further processing effectiveness and future works could compare the other state-of-the-art segmentation algorithms with results of multi-resolution algorithm.