×
نام و نام خانوادگی
بازخوانی ...
اطلاعات کتابشناختی
عنوان اصلی: بهبود طبقه‌بندي داده‌هاي سري زماني سنجش از دوري با استفاده از الگوريتم‌هاي يادگيري كرنل چندگانه و فعال
عنوان: Improvement of classification algorithms for remotely sensed time series data using multiple kernel and active learning algorithms
پدیدآورندگان : سعيد (پديدآور)
نيازمردي (پديدآور)
عبدالرضا (پديدآور)
صفري (پديدآور)
رشته مهندسي نقشه برداري-پرديس دانشكده هاي فني (پديدآور)
نوع : متن
جنس : پايان نامه
صاحب محتوا :

کتابخانه دیجیتالی دانشگاه تهران

وضعیت نشر :
خلاصه : Nowadays, Satellite Image Time-Series (SITS) data are widely available. SITS data are a set of images acquired over the same geographical area at different times and have different applications in various fields. Crop mapping through classification of SITS data is one of these applications which has gained a lot of attention due to its important role of crops in human food basket. SITS data can provide more information regarding the dynamic spectral behaviors of crops during the course of time in comparison with a single-time image. The SITS data that contained the images acquired by multispectral or hyperspectral sensors can be considered as multivariate SITS. Crop mapping through multivariate SITS data classification is a very challenging task due to three main reasons. First, this type of SITS data is a four-dimensional data, which cannot be classified using the conventional classification algorithms. Second, the same class may have very different statistical characteristics as a result of the variation of the spectral behavior of crops and the changes in the atmosphere and sensor condition over the time. Finally, for multivariate SITS data classification, adequate number of training samples which are informative in all the images of the SITS data are required. The purpose of this dissertation is to propose algorithms for addressing all these issues. To this end, we proposed multiple kernel representations for classifying multivariate SITS data. These representations initially construct a number of kernels from the most informative parts of the time-series data, such as different images of the time-series and the changes in the trend of crops’ spectral reflectance. Then, multiple kernel learning (MKL) algorithms are used to optimally combine these kernels into a composite kernel. In here, three multiple kernel representations are presented, that differ from each other considering the way that they construct their corresponding kernels. In our experiments, we evaluated the performances of these representations by using several MKL algorithms. In order to address the problems associated with the availability of training samples, we proposed multiple kernel active learning algorithm. This interactive algorithm uses both active and multiple kernel learning algorithms to select the most informative samples for classification of time-series data. In this dissertation, we evaluated the performances of the proposed algorithms for classification of the time-series consisted of the multispectral images acquired by the RapidEye and SPOT sensors. We also proposed new MKL algorithms and compared their performances with the conventional MKL algorithms. The obtained results of experiments showed an increase in the classification accuracy up to 18% by using multiple kernel representations. In addition, the accuracies more than 85% were achieved in the case of using the samples obtained from the multiple kernel active learning algorithms, which only required 2 seconds for its computation. Keywords: Multivariate time-series, classification, Multiple Kernel Learning, Active Learning.
یادداشت :
كتابنامه: به انگليسي
چكيده: به فارسي و انگليسي
گرايش سنجش از راه دور
دكتري
شناسه : oai:ut.ac.ir:thesis/1-312541
تاریخ ایجاد رکورد : 1396/5/31
تاریخ تغییر رکورد : 1396/7/4
قیمت شيء دیجیتال : دارای قیمت

راهنمای سفارش دیجیتال سازی

* محتوای این صفحه توسط کارشناسان این درگاه ویرایش نشده است. لطفا در صورت مشاهده ایراد در محتوا از این طریق اطلاع رسانی کنید.

دیدگاه شما

تست
ورود به درگاه کنسرسیوم
Loding



رمز عبور خود را فراموش کرده ام.
چنانچه تا کنون عضو سایت نشده اید ثبت نام کنید.
درباره کنسرسیوم
ما مجموعه‌ای از كتابخانه‌ها و سازمان‌های دارای منابع اطلاعاتی (کتاب، نشریه، نسخه‌های خطی، عکس، صدا، فیلم و... ) هستیم که با هدف تامین نیازهای پژوهشگران و شهروندان ایرانی برای دسترسی هر چه سریع‌تر به محتوای مورد نظر خود، کنسرسیوم محتوای ملی را تشکیل داده‌ایم. برای رسیدن به این هدف، قصد داریم با بسترسازی مناسب و جلب مشارکت دیگر تولید کنندگان محتوا به گرد آوری، تبدیل، سازماندهی و حفاظت اطلاعات به شکل رقومی و در سطح ملی، بپردازیم.