کد مقاله | کد نشریه | سال انتشار | مقاله انگلیسی | نسخه تمام متن |
---|---|---|---|---|
4972830 | 1451242 | 2017 | 13 صفحه PDF | دانلود رایگان |
عنوان انگلیسی مقاله ISI
Contextually guided very-high-resolution imagery classification with semantic segments
ترجمه فارسی عنوان
طبقه بندی تصاویر با وضوح با وضوح بالا با بخش های معنایی به صورت محتوا هدایت می شود
دانلود مقاله + سفارش ترجمه
دانلود مقاله ISI انگلیسی
رایگان برای ایرانیان
کلمات کلیدی
موضوعات مرتبط
مهندسی و علوم پایه
مهندسی کامپیوتر
سیستم های اطلاعاتی
چکیده انگلیسی
Contextual information, revealing relationships and dependencies between image objects, is one of the most important information for the successful interpretation of very-high-resolution (VHR) remote sensing imagery. Over the last decade, geographic object-based image analysis (GEOBIA) technique has been widely used to first divide images into homogeneous parts, and then to assign semantic labels according to the properties of image segments. However, due to the complexity and heterogeneity of VHR images, segments without semantic labels (i.e., semantic-free segments) generated with low-level features often fail to represent geographic entities (such as building roofs usually be partitioned into chimney/antenna/shadow parts). As a result, it is hard to capture contextual information across geographic entities when using semantic-free segments. In contrast to low-level features, “deep” features can be used to build robust segments with accurate labels (i.e., semantic segments) in order to represent geographic entities at higher levels. Based on these semantic segments, semantic graphs can be constructed to capture contextual information in VHR images. In this paper, semantic segments were first explored with convolutional neural networks (CNN) and a conditional random field (CRF) model was then applied to model the contextual information between semantic segments. Experimental results on two challenging VHR datasets (i.e., the Vaihingen and Beijing scenes) indicate that the proposed method is an improvement over existing image classification techniques in classification performance (overall accuracy ranges from 82% to 96%).
ناشر
Database: Elsevier - ScienceDirect (ساینس دایرکت)
Journal: ISPRS Journal of Photogrammetry and Remote Sensing - Volume 132, October 2017, Pages 48-60
Journal: ISPRS Journal of Photogrammetry and Remote Sensing - Volume 132, October 2017, Pages 48-60
نویسندگان
Wenzhi Zhao, Shihong Du, Qiao Wang, William J. Emery,