Image-based Determination of Plum “Tsuyuakane” Ripeness via Deep Learning

DOI Web Site 2 Citations 5 References
  • Tatemoto Satoshi
    Agricultural and Horticultural Research Division, Tokushima Agriculture, Forestry and Fisheries Technology Support Center
  • Harada Yoko
    Agricultural and Horticultural Research Division, Tokushima Agriculture, Forestry and Fisheries Technology Support Center
  • Imai Kenji
    Resources and Environmental Research Division, Tokushima Agriculture, Forestry and Fisheries Technology Support Center

Bibliographic Information

Other Title
  • 深層学習を利用したウメ「露茜」の画像による熟度分類

Search this article

Description

<p>We determined the ripeness of plums (“Tsuyuakane”) from images with combined object detection by using a single-shot multibox detector (SSD) and a convolutional neural network (CNN). From June to July 2018 we obtained digital images and movies to characterize changes in the fruits on Tsuyuakane trees: 443 images were used to train the SSD in fruit recognition, and 5823 images were used to train the CNN for ripeness classification. The pictures were subsequently classified into five groups based on fruit ripeness, with the resulting classification accuracy at 94%. Subsequent training of the CNN with 366 images originally withheld from SSD training resulted in an increase of ripeness classification accuracy to 96%. These results suggest that this deep-learning method can be utilized to determine fruit ripeness based on training with sequential images depicting fruit growth.</p>

Journal

Citations (2)*help

See more

References(5)*help

See more

Keywords

Details 詳細情報について

Report a problem

Back to top