مقایسه عملکرد شبکه عصبی مصنوعی، ماشین بردار پشتیبان و مدل شیءگرا در پایش تغییرات سطح پوشش برف با استفاده از تصاویر چند زمانه لندست (مطالعه موردی: کوهستان الوند)

نوع مقاله: مقاله پژوهشی

نویسندگان

1 دانشگاه پیام نور

2 دانشگاه تبریز

3 اداره کل هواشناسی استان همدان

4 اداره کل هواشناسی استان همدان

5 شرکت آب منطقه‌ای استان همدان

چکیده

پوشش برف و تغییرات زمانی آن، از پارامترهای اساسی در بررسی­های هیدرولوژیکی و اقلیم شناسی می­باشند. امروزه با استفاده از تصاویر ماهواره­ای می­توان به ارزیابی تغییرات سطح پوشش برف در سری­های زمانی مختلف پرداخت. پژوهش حاضر با هدف پایش تغییرات سطح پوشش برف در کوهستان الوند همدان با استفاده از داده­های رقومی ماهواره لندست در سری­های زمانی سال­های 1975، 1986، 1993، 2001، 2008 و 2018 انجام گرفته است. روش تحقیق در این پژوهش، استفاده از طبقه­بندی شبکه عصبی مصنوعی، ماشین بردار پشتیبان و مدل شیء گرا جهت برآورد سطح پوشش برف بوده است که پس از انجام عملیات پیش پردازش بر روی تصاویر ماهواره­ای، نقشه­های طبقه­بندی سطح پوشش برف کوهستان الوند از روش­های شبکه عصبی، ماشین بردار پشتیبان و مدل شیءگرا تهیه گردید. سپس صحّت این روش­ها مورد ارزیابی قرار گرفت. پژوهش حاضر نشان داد به ترتیب، مدل شیءگرا، ماشین بردار پشتیبان و شبکه عصبی مصنوعی دارای بالاترین میزان دقت بودند، لذا تغییرات مساحت سطح پوشش برف در سری­های زمانی مختلف، با استفاده از روش شیءگرا محاسبه گردید. مساحت بدست آمده برای سطح پوشش برف در کوهستان الوند با استفاده از مدل شیءگرا به ترتیب عبارت بودند از، سال 1975 برابر با 630 کیلومتر مربع، سال 1986 برابر با 611 کیلومتر مربع، سال 1993 برابر با 414 کیلومتر مربع، سال 2001 برابر با 151 کیلومتر مربع، سال 2008 برابر با 242 کیلومتر مربع و سال 2018 برابر با 154 کیلومتر مربع که نشانگر کاهش چشمگیر سطح پوشش برف از سال 1975 تا سال 2018 در کوهستان الوند می­باشد.

کلیدواژه‌ها


عنوان مقاله [English]

Comparison performance of artificial neural network, support vector machine and object-oriented model for monitoring snow cover surface changes using Landsat multi temporal images (Case study: Alvand mountain)

نویسندگان [English]

  • Mostafa MousaPour 1
  • Bakhtiar Feizizadeh 2
  • Syed Akbar Hosseini 3
  • Hasan Kerchi 4
  • Seifi Azadeh 5
1 Payam Nour University
2 Tabriz University
3 Meteorological Administration of Hamedan Province
4 Meteorology of Hamadan Province
5 Water resource expert, Regional Water resource of Hamadan province.
چکیده [English]

Expanded abstract:
 snow is one of the most important forms of precipitation in hydrology cycle in mountainous basin which plays an important role on agricultural and domestic water supply resources as delayed flows in high flow seasons and minimal flow in low flow seasons and energy production. today, the use of remote sensing data is applied at obtaining the area of accurate snow cover data in the efficient management of water resources. The purpose of this study is to determine the changes in snow cover in alvand mountains of hamedan using of remote sensing. The research method is using of artificial neural network classification, support vector machine, and object oriented model, that with using of the most appropriate method among them has been calculated, the amount of snow cover area variations in different time series. Alvand mountain in hamadan province is located between hamedan, tuyserkan, asadabad and bahar. Its highest mountaintop, called alvand, is located 18 kilometers south of hamadan city and is 3584 meters above sea level. The direction of this mountain is drawn from the northwest to the southeast and the hamedan province divides into two northern and southern halves. The data used in this study include sensor images MSS, TM, ETM +, and OLI Landsat satellite in the time series of 1975, 1986, 1993, 2001, 2008, and 2018. To prepare a map of changes the snow cover area, was carried out processing operations on satellite images in three stages: preprocessing, processing, and post processing. Similar spectral separation and division of the class which has the same spectral behavior are called satellite information classification. The main purpose of classification of digital images is to create subject maps. In recent years, new approaches have been proposed to concurrent with the advancement of image computer processing technology, for example, the use of neural networks, tree decisions, and methods derived from fuzzy logic theory, the use of secondary information such as texture, background and ground effects are the most important of these methods. An artificial neural network algorithm is a method in the field of machine learning and artificial intelligence that eventuates from the human nervous system to analyze complex nonlinear models and parallel computing systems. One of the advantages of artificial neural networks is that they are independent of the assumption of statistical distribution. Neural networks are nonlinear and can transform the input data into a desired output as a complex mathematical function. Support vector machine is a sample classification method that first time was introduced by Vladimir vapnik. This method is a non-parametric supervised statistical method to classify the classes in the training data, super surface practice on them. The support vector machine is one of the supervised classification algorithms that predict every sample stand in which class or group. This algorithm has less sensitive to the phenomena of multidimensional space, for this reason, it is a suitable method for the classification of multi-spectral and hyperspectral data. One of the advantages of a support vector machine algorithm is to provide a good classified image resolution with small training samples. In recent years, many research has been carried out on the applications of fuzzy logic in remote sensing, have largely been based on object oriented methods, in addition to numerical values, is used the data of texture, shape and tone color, in classification process. The ability to classify the base pixels method is limited when different ground objects are recorded with the same numeric values on a digital image. The object oriented classification method has been proposed to solve this problem. One of the clearest difference between the basic image pixel processing and object oriented image processing are in processing of object oriented image, the processing basic units are image objects or segments, not single pixels, the other difference is that the classification in object-oriented image processing is soft classification, which is based on fuzzy logic. After operation preprocessing on satellite images, maps of classification the snow cover area was provided of this three mentioned method from alvand mountain. Then the validity of these methods was evaluated. This research specified that object oriented model, support vector machines and artificial neural network have the highest accuracy respectively and thus changes of snow cover area were calculated in different time series using the object-oriented method. The snow cover area obtained using of the object-oriented model were 630, 611, 414, 151, 242, 154 square kilometers in 1975, 1986, 1993, 2001 ,2008 and 2018 respectively, indicating the area of the snow cover have diminished significantly from 1975 to 2018 in the alvand mountains.

کلیدواژه‌ها [English]

  • Remote Sensing
  • Artificial Neural Network
  • Support Vector Machine
  • object base
  • alvand mountain