نوع مقاله : مقاله پژوهشی

نویسندگان

• فاطمه احمدی ¹
• یاسر ابراهیمیان قاجاری ²
• عباس کیانی ²

¹ دانشجوی کارشناسی ارشد فتوگرامتری، دانشگاه صنعتی نوشیروانی بابل

² استادیار دانشکده عمران، دانشگاه صنعتی نوشیروانی بابل

چکیده

گسترش شهرها یک فرآیند پویاست و به‌روزرسانی نقشه‌های شهری به‌منظور ارائه خدمات بسیار مهم است. داده­های سنجش‌ازدوری منبع قدرتمندی برای استخراج مناطق ساخته‌شده می­باشند؛ یکی از داده‌هایی که می‌تواند پویایی مناطق شهری را تشخیص دهد داده‌های نور شب است. ازآنجایی‌که مناطق شهری در هنگام شب توسط نورهای مصنوعی موجود در منازل و خیابان‌ها روشن می‌شوند، درنتیجه به­خوبی از پس­زمینه متمایز خواهند شد. ازاین­روی مطالعه حاضر در تلاش است رشد و گسترش مناطق شهری را با استفاده از نمونه‌های آموزشی با کیفیت و اتوماتیک از روی ترکیبی از تصاویر نور شب و اپتیک، در یک بازه 24 ساله با دقت بالایی شناسایی و استخراج کند. دراین­راستا برای تولید نمونه‌های آموزشی با کیفیت، شاخص نور شب تحت عنوان [1]VTNUI توسعه داده شده است که با ترکیب ویژگی‌های مختلف بدست آمده از تصاویر لندست و نور شب در محدوده‌های شهری و در نظر گرفتن روابط بین مناطق، پدیده اشباع[2] و شکوفایی[3] تصاویر نور شب در محدوده شهری را کاهش داد. سپس با بررسی حد آستانه‌های خودکار بر روی شاخص توسعه داده شده نمونه‌های آموزشی با کیفیت تولید شد تا طبقه‌بندی دقیق‌تری از مناطق شهری ارائه شود. ازاین‌روی ابتدا از نمونه‌های آموزشی اولیه به دست آمده با استفاده از حد آستانه خودکار بر روی تصاویر نور شب طبقه‌بندی اولیه صورت پذیرفت سپس نمونه‌ها با اعمال حد آستانه خودکار بر روی شاخص معرفی شده پالایش شدند و طبقه‌بندی نهایی صورت گرفت. درنهایت بر اساس تحلیل سری زمانی، روند رشد منطقه بررسی شده است. به منظور بررسی اثربخشی روش پیشنهادی، دو منطقه دارای اقلیم متفاوت انتخاب شد و بررسی­های مختلف بصری و کمی برای ارزیابی صورت پذیرفت. نتایج طبقه‌بندی نهایی برای بابل و کرمان به ترتیب با میانگین ضریب کاپا 0/93 و 0/74 و میانگین دقت کلی 97/76 و 87/63 برای تمام سال‌های مورد بررسی به دست آمده است.
 
[1] vegetation and Temperature -NTL urban index
[2] saturation
[3] blooming

کلیدواژه‌ها

• داده نور شب
• اشباع و شکوفایی
• پویایی گسترش شهری
• تصاویر ماهواره ای
• نمونه‌های آموزشی خودکار

موضوعات

• استخراج، پردازش، تولید و نمایش داده های جغرافیایی

عنوان مقاله [English]

Investigating the dynamics and expansion of urban areas using automatic training samples obtained from the combination of night light and medium resolution satellite imagery

نویسندگان [English]

• Fatemeh Ahmadi ¹
• Yasser Ebrahimian Ghajari ²
• Abbas Kiani ²

¹ MSc student in photogrammetry , Babol Noshirvani University of Technology, Mazandaran, Iran

² Assistant Professor, Dept. of Geomatics, Faculty of Civil Engineering, Babol Noshirvani University of Technology, Mazandaran, Iran

چکیده [English]

Extended Abstract
Introduction
Urbanization can be called as a progress in the social and economic field, therefore, urban planners need timely information to provide services and urban management. Since traditional methods are difficult; The use of automatic and semi-automatic methods has become highly necessary; Therefore, remote sensing data and remote sensing image classification techniques have been used to help identify land use. One of the images that can effectively identify human activities and urban areas is night light data. Unlike other satellites, these satellites image the surface of the earth at night and can clearly separate urban areas from the surrounding areas that are off during the night. With these interpretations, it can be said that for a region there are various data with different nature and capabilities (spectral, night light, etc.); that each of these data has advantages and limitations, and the combined use of these data will bring the possibility of increasing accuracy and reducing uncertainty, in this regard, algorithms and scientific methods that enable the combination of these data are of great importance. This study, using a combination of night light images and multispectral images taken during the day, tries to specifically meet the main goal of the research, which is the extraction of built-up areas, by producing automatic and high-quality optimal training samples.
Material & Methods
In order to evaluate the results, the two study areas of Babol and Kerman with two different climates have been investigated. Also, DMSP and VIIRS night light images and Landsat 5, 7 and 8 images from the same years have been used.
Research Methods
The proposed approach in the current research includes four main phases of pre-processing, feature extraction and production of initial training samples, selection of optimal training samples and finally classification and evaluation. Therefore, firstly, night light images are corrected. Then, by using the limit of automatic thresholds on night light images, primary samples are produced, which include two classes of built and unbuilt areas; Since night light data is related by human activities, built-up areas usually have higher night-light values while non-built areas have lower or zero values. On the other, due to saturation and blooming problems in DMSP images and the relatively low spatial resolution of night light data compared to Landsat images, the training samples of built areas using night light data still include non-built areas such as water and vegetation. Therefore, an index has been developed based on the features of night light images and the features extracted from Landsat images. in which it has been tried by integrating and considering the inverse relationship between the features of urban areas and other areas such as vegetation and soil in  images The land surface temperature (LST) obtained from the thermal band of the Landsat image and the NDVI vegetation index obtained from the Landsat and the characteristics of urban areas in the night light image provide an index that, while maintaining the main characteristics of urban areas in the night light images, the saturation  and blooming is minimize , the final classification done by optimizing  initial training samples.
Result & Discussion
After correcting the night light images, an upward trend was shown for the pixel values of each city, which indicates the correctness of the pre-processing. Then, based on the characteristics of each of the images, the appropriate automatic threshold limit was applied on the night light images to produce the initial training samples. In the following, the night light images are corrected by the introduced index to minimize the saturation and blooming in the urban areas and its suburbs, and with the training samples optimized by this process, the final classification operation is done. the results indicated that due to the quality  the initial training samples, the classified pixels related to urban areas have been obtained more than the reality and this condition was such that in Kerman city in some years the classification was lost and practically no classification was done and this shows the low quality of the initial training samples; Because in the inital samples, due to the low spatial resolution of the night light images, the size of the samples related to the built-up areas was detected to be large, and in this way, there are definitely samples in the specified range that are related to vegetation, soil, etc. In the next step, the classification is done using optimal training samples, in which the regions created in this mode were modified and are closer to the reference data and reality; In fact, the reason for this is that when night light and Landsat data are used in combination, they overcome each other's limitations.
Conclusion
The most important challenge in the topic of classification is the selection of training samples because a main and fundamental pillar in classification is training samples, and if there is a valid training sample, the classification is done with precision, and inasmuch as obtaining training samples manually is costly and time-consuming. There is Feeling need to obtain training samples automatically. Therefore, one of the main goals of this study is to classify and extract built-up areas using the potential of satellite images; And the initial training samples can be obtained automatically from the night light image, but due to the high saturation and blooming of the night light images, these samples do not have high quality. therefore, to solve this problem, a night light index has been developed, which takes into account the relationship Between the characteristics of the city in optical images and night light, it has been able to minimize the problems of this data to a great extent in both studied areas that had two different climates ,which shows its flexibility and effectiveness, and in this way, high-quality educational samples with a limit automatic thresholds were obtained from this index, which were very effective in the final classification process. the examination of the time series of the growth of cities showed that most of the growth and expansion happened around the city.

کلیدواژه‌ها [English]

• Night light data
• saturation and blooming
• automatic training samples
• urban built-up areas