動的輪郭モデルを用いた植物と画像背景の分離および転移学習による植物分類

DOI Web Site

書誌事項

タイトル別名
  • Background and Foreground Segmentation in Plant Images with Active Contour Model and Plant Image Classification using Transfer Learning
  • ドウテキ リンカク モデル オ モチイタ ショクブツ ト ガゾウ ハイケイ ノ ブンリ オヨビ テンイ ガクシュウ ニ ヨル ショクブツ ブンルイ

この論文をさがす

抄録

     Distribution of natural vegetation is closely related to its biodiversity. Precise investigation and evaluation of the distribution is important. An automatic classification method of natural plant images taken by cameras will enable the field research by non-professionals and lead to its high efficiency. To date, few researches are available on the automatic classification using transfer learning, which is a kind of convolutional neural network (CNN) representing high performance in image classification. For the preparation of the training data sets, generally lots of images, at least from hundreds to thousands, are necessary. It is expected that background segmentation in the images will contribute to higher classification accuracy. It is very laborious and time-consuming to segment the background manually one by one. From this reason, the background segmentation was automatically conducted with active contour model which is widely used method to extract the border line of an object in images. However, whether the segmentation leads to high classification accuracy of the method using CNN has not been discussed enough. In this research, plant image classification with transfer learning was conducted with plant images whose background and foreground were segmented and not segmented. We found that the classification accuracy was about 99% and the segmentation was not necessarily needed for the classification.

収録刊行物

  • Eco-Engineering

    Eco-Engineering 30 (3), 81-85, 2018-07-31

    生態工学会

関連プロジェクト

もっと見る

キーワード

詳細情報 詳細情報について

問題の指摘

ページトップへ