Automatic position correction using center estimation for cereal images

We use class activation maps from cereal species classification to estimate the center of a set of spikelets. This estimated center is used to automatically correct positional misalignment introduced when acquiring cereal images.

To record how different cereal species look, we capture images of their spikes using a scanner. The position of a spike in the image can vary depending on how it is placed and how the scanner is used. This research aims to correct these shifts automatically.

We use a class activation map, which shows the image regions emphasized by a computer when it identifies the cereal species. From this map, we estimate the center of the cluster of spikelets, the small units that make up a spike. We then align the image using this center. Experiments confirmed that the method could accurately locate the center of the spikelet cluster and correct the images' positional misalignment.

• Kotarou Saku, Takumi Ozaki, Shun Sakuma, Michiko Inoue, and Masashi Nishiyama,
  Automatic position correction using center estimation for cereal images, Proceedings of IEEE 13th Global Conference on Consumer Electronics (GCCE), POS, pp. 492 - 495, October 2024. [Poster]
• 佐久 高太郎, 尾崎 匠, 佐久間 俊, 井上 路子, 西山 正志,
  ムギ画像の位置補正に向けた種分類のクラス活性化マップの利用手法の提案,
  画像センシングシンポジウム (SSII), IS1-20, pp. 1 - 7, June 2024. [Poster] [Short movie]
• 佐久 高太郎，佐久間 俊，井上 路子，西山 正志,
  ムギの中心位置推定を用いた穂画像の自動位置調整,
  ビジョン技術の実利用ワークショップ (ViEW), OS3-H4(IS3-4), pp. 352 - 358, December 2023. [Slides] [Short movie]