Deep Learning-Based Classification of Crowdsourced Typhoon Disaster Images

• Main Authors: CHEN, I-CHING, 陳怡靜
• Other Authors: CHU, TSUNG-HSIEN
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/v5g4eb

碩士 === 國立雲林科技大學 === 資訊工程系 === 107 === Due to the advance of Internet and Web 2.0 technologies, social media like Facebook, Line, and Twitter, has played an important role in disaster response, especially for large-scale disasters. Although social media can be a valuable source of real-time information during disasters, the crowdsourced information should be verified before it can be used to support rescue plans. In this paper, we designed a deep-learning based validation system, named CTIC, to classify crowdsourced typhoon disaster information. A convolution neural network is used to categorize social-media-based disaster information into four different types: fallen-sign, fallen-tree, broken-road and flood. In our experiment, for each category, we used 40 photos for training and 40 photos for testing. Our results show that the average accuracy rate of CTIC is 80.8%. In particular, for the category of fallen-sign, the accuracy rate is up to 91.9%.