Flora and fauna identification system based on multi-modal data observed

2017 Fiscal Year Final Research Report

• Project/Area Number: 15K12027
• Research Category: Grant-in-Aid for Challenging Exploratory Research
• Allocation Type: Multi-year Fund
• Research Field: Multimedia database
• Research Institution: Toyohashi University of Technology
• Principal Investigator: Aono Masaki 豊橋技術科学大学, 工学(系)研究科(研究院), 教授 (00372540)
• Project Period (FY): 2015-04-01 – 2018-03-31
• Keywords: 植物鑑定 / 深層学習 / 画像特徴量

Outline of Final Research Achievements

From the plant photo images posted on the Internet, We have participated in the international contest called PlantCLEF from the second year of the research period, resulting in No.1 world record in 2016 and the year 2017, "Noisy training data" truck, which was the second place as a whole. Technically, we have succeeded in improving the accuracy by mixing textual features from meta data with the image features of CNN and MLP layers of the deep learning as well as the traditional features such as Fisher vector. Furthermore, we have developed a fractional pooling method to enable convolution with arbitrary rectangular images.

Free Research Field

マルチメディア　データマイニング