「海底の地層断面の画像データによる資源探査への機械学習の応用」の説明図
![](http://mathsci.math.akita-u.ac.jp/wp-content/uploads/2017/09/cores-300x35.png) Original image of a core (top) and the manually edited one, masking the ruler and secondary color artifacts with the alpha channel (bottom) |
![](http://mathsci.math.akita-u.ac.jp/wp-content/uploads/2017/09/FCNmodel-331x1024.png) The Fully Convolutional Neural Network (FCN) used for segmentation (from [Long, Shelhamer, Darrell, 2014])
|
![](http://mathsci.math.akita-u.ac.jp/wp-content/uploads/2017/09/IoU-300x211.png) Intersection-over-Union (IoU) is used to compare the performance of the neural net to the manual editing |
![](http://mathsci.math.akita-u.ac.jp/wp-content/uploads/2017/09/otsu-300x133.png) Otsu’s binary thresholding method for turning gray scale images into black-and-white ones |
![](http://mathsci.math.akita-u.ac.jp/wp-content/uploads/2017/09/pred1-300x98.png)
Images of the original core
![](http://mathsci.math.akita-u.ac.jp/wp-content/uploads/2017/09/pred2-300x99.png)
manually edited
![](http://mathsci.math.akita-u.ac.jp/wp-content/uploads/2017/09/pred3-300x100.png)
prediction of the FCN