回転不変特徴量を用いた階層型ニューラルネットワークによる魚種選別システムの開発

DOI Web Site Web Site 参考文献21件

書誌事項

タイトル別名
  • Development of Sorting System for Fishes by Feed-forward Neural Networks Using Rotation Invariant Features
  • カイテン フヘン トクチョウリョウ オ モチイタ カイソウガタ ニューラル ネットワーク ニ ヨル ギョシュ センベツ システム ノ カイハツ

この論文をさがす

抄録

In this research, we have developed a sorting system for fishes, which is comprised of a conveyance part, a capturing image part, and a sorting part. In the conveyance part, we have developed an independent conveyance system in order to separate one fish from an intertwined group of fishes. After the image of the separated fish is captured in the capturing part, a rotation invariant feature is extracted using two-dimensional fast Fourier transform, which is the mean value of the power spectrum with the same distance from the origin in the spectrum field. After that, the fishes are classified by three-layered feed-forward neural networks. The experimental results show that the developed system classifies three kinds of fishes captured in various angles with the classification ratio of 98.95% for 1044 captured images of five fishes. The other experimental results show the classification ratio of 90.7% for 300 fishes by 10-fold cross validation method.

収録刊行物

参考文献 (21)*注記

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