Aerial Image Semantic Segmentation Using Neural Search Network Architecture

• Published in: Multi-Disciplinary Trends in Artificial Intelligence Vol. 11248; pp. 113 - 124
• Main Authors: Bui, Duc-Thinh, Tran, Trong-Dat, Nguyen, Thi-Thuy, Tran, Quoc-Long, Nguyen, Do-Van
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2018; Springer International Publishing
• Series: Lecture Notes in Computer Science
• Subjects: Aerial image; Neural Search Network Architecture; Semantic segmentation
• ISBN: 303003013X; 9783030030131
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-030-03014-8_10

In remote sensing data analysis and computer vision, aerial image segmentation is a crucial research topic, which has many applications in environmental and urban planning. Recently, deep learning is using to tackle many computer vision problem, including aerial image segmentation. Results have shown that deep learning gains much higher accuracy than other methods on many benchmark data sets. In this work, we propose a neural network called NASNet-FCN, which based on Fully Convolutional Network - a frame work for solving semantic segmentation problem and image feature extractor derived from state-of-the-art object recognition network called Neural Search Network Architecture. Our networks are trained and judged by using benchmark dataset from ISPRS Vaihingen challenge. Results show that our methods achieved state-of-the-art accuracy with potential improvements.