Skip to content

Identification of images containing invasive species using convolutional neural networks

Notifications You must be signed in to change notification settings

jw15/kaggle_invasive_species

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

16 Commits
 
 
 
 
 
 

Repository files navigation

Kaggle Competition: Invasive Species

Identifying invasive hyacinth flowers in images of forest scenes using computer vision (competition link)

First attempt: convolutional neural net (Keras with Theano backend) using images resized to 150 x 200 pixels: 68% accuracy in 2 epochs. Very slow with the larger images. Definitely room for improvement.

Second attempt (code is in /src): CNN using images centered and resized to square (256x256), then cropped to 224x224. Used keras image generator to generate images that are rotated, random height/width shifts, flipped on horizontal axis). Trained neural net on top of VGG16. 98% accuracy in 40 epochs. This code was adapted from (work by kaggle user fujisan).

Things to try next:

  1. Train on larger images (Using images resized to 512x512 and cropped to 488x488 so far is not working very well.)
  2. Try multiscale model
  3. Identify regions/partitions of the image where flowers are located

About

Identification of images containing invasive species using convolutional neural networks

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages