Animal Classification

About the Dataset

DATASET: https://docs.google.com/forms/d/e/1FAIpQLSeerwGmVkxbZsI0jxGLWfND8nmRCDDSoW1OAVpa8ZzCCR2e-A/viewform

1. Meta information

   • 50,000 Images
   • resolution: 64x64 (RGB)

2. Noise Rate

   • The noise rate was estimated at 8%.

3. File name & Label Information

   • file name is "label" + _ + "indexed name"
   • label mappings
     • 0: cat
     • 1: lynx
     • 2: wolf
     • 3: coyote
     • 4: cheetah
     • 5: jaguar
     • 6: chimpanzee
     • 7: orangutan
     • 8: hamster
     • 9: guinea pig

About the Model

• A VGG19 Computer-vision model that can classify images of animals into the 10 categories mentioned above

Uses a large size dataset of 50,000 images with around 5000 images of each class to train the model. The images were then split into training and validation data: 80% of the data was used for training and the remaining 20% was used for testing.

Data Modeling

My input data consisted of 40000 RGB 64x64 images. Given a directory containing all the images, I utilize Keras' 'ImageDataGenerator' to create a training The list containing data variables is then normalized and both label and data lists are re-assigned as numpy arrays. After the data is split up by category and divided into training and testing, I used Keras' ImageDataGenerator to create a testing and training, and used to perform some transformations on the data, such as zooming or rotating the image, to make it more impactful when training.

Architecture:

Limitations

Possible Improvements

• Another CNN for animaml recognition or an effective algorithm for preprocessing the data and forming a 'region of interest' around each animal
  • Add another neutral net to create a region of interest around the target in the image