استخراج محدوده ساختمانها از تصاویر ماهواره ای نه تنها نقش انسان را در تولید نقشههای شهری کاهش میدهد بلکه در زمان و هزینه تولید این نقشهها صرفه جویی قابل توجهی صورت خواهد گرفت. همچنین امکان تهیه نقشه محدوده ساختمانها در بازههای زمانی کوتاه مدت سالیانه، ماهیانه و حتی کوتاه تر را نیز فراهم میآورد. شهرداریها و سایر ارگانهای شهری نیاز به نقشههای بروز از ساختمانها و سایر عوارض شهر دارند. روش انتخابی برای انجام این کار بر اساس اطلاعات هر پیکسل(Pixel Base) و با استفاده از الگوریتم درخت تصمیم گیری (Decesion Tree) است. به این صورت که با تعیین حد آستانههای مناسب با توجه به ویژگیهای طیفی هر پدیده نسبت به حذف مرحله ای عوارض اقدام و در نهایت کلاس ساختمانها با توجه به هدف مطالعه استخراج گردید. عملیات فوق بر روی تصاویر دو زمانه مربوط به سالهای ۲۰۰۸ و ۲۰۱۲میلادی انجام و سپس نقشه تغییرات ساختمانها استخراج گردید و درنهایت نقشه خروجی با نقشههای رقومی املاک در وضع موجود که از ممیزی سطح شهر تولید شده مقایسه و دقت روش انتخابی مورد ارزیابی قرار گرفته است.
عنوان مقاله [English]
Building detection from world view images using decision tree algorithm
Building detection by satellite images decrease not onlyhuman role in city map production but also time and cost for those map production this also provide the opportunity for building map production in monthly and yearly periods. The mayoralty and all city organizations need an updated building maps.
In the past, building detection from digital images was done manually and as a result, this process which was done slowly and costly, needs professional operator. This kind of detection was not appropriate for city area with high intensity.today, high resolution satellite images areprovided using remote sensing technology development. By use of these images, we are capable of Land use classifyingin city area.
In high measure IKONOSSatellite was known as the first high separation commercial satellite in 1999.Afterthat similar satellites such as and quick bird geoeye werelaunched. Using a suitable automaticOr semi-Automatic method for building detection not only reducehuman role in city map production but also reduce cost and time for the map production and provide us with an opportunity to produce building maps in monthly and yearly periods.In addition,this method is useful for other objects with similar nature and geometry.Nowadays, mayoralties and other organizations which need updated information related to domains and city alley undergo many coststo survey domains by difficult operations and many professionals, most of the time, the accuracy of information was not satisfying and this lead to repetition.
In this research,multi spectral and panchromatic images of band1 to 4 of quick bird satellite were used to obtain city object Class boundary such as buildings,roads,plants, etc.
The method is based on pixel base information using decision tree algorithm,i.e,by determiningSuitable there should level based on each phenomenon spectral characteristics,objects were deletedand finally building classes were detected based on study purpose.
The above mentioned activates were done on images related to 2008 and 2012 and then building change maps were detected,finally,the output map was compared with domain digital maps By city survey and the method accuracy was evaluated.
â¢ By checking phenomenon observed in chabaharr satellite images,Decision tree for separation of following objects was chosen:
â¢ Plants,park and vegetations
â¢ Black top of street and ally surface
â¢ Building back top
â¢ Building and tree shadows which are observable because of time of recording
â¢ pavementcovered by dust which are part of wasteland,
â¢ Main buildings produced by other phenomenon separation.
After that, related phenomenon in each class of 2008 and 2012 images were separated and change percentage was calculated building courage percentage in 2008 image was 16.41 which was increased to 17.57 because of black tap percentage of 1.16 this percentage was increased to 24.69 in 2012 image i.e. we had the increase of 7.12 percentage as a result of new construction in the period of 4 years.
After accuracy evaluation using survey experimented information and obtaining building parcels diversity in 2 layers , we observed the diversity about 8.2 percentage which is explainable this diversity can be because of different errors of measurement such as , using map whit 2000/1 measure as a base map for survey , errors during detection without reduction unbuilt area such as in survey map .Therefore, this method can be usual for new building detection without local reference.