metadata

license: apache-2.0
language:
  - en
metrics:
  - accuracy
library_name: keras
pipeline_tag: image-classification
tags:
  - astronomy

Model Card for Model ID

This model classifies RGB images to the 2 classes, Spheroid or Spiral.

Model Details

Model Description

Developed by: Jeroen den Otter
Funded by: NASA
Shared by [optional]: Michael Rutkowski
Model type: Keras Sequential
Language(s) (NLP): English
License: Apache2

Model Sources [optional]

Repository: https://www.kaggle.com/c/galaxy-zoo-the-galaxy-challenge/overview
Paper [optional]: In progress

Uses

The model can be used for identifying different galaxies from cutout images. It does not provide bounding boxes, so multiple galaxies in 1 image is not desired.

How to Get Started with the Model

Use the code below to get started with the model.

model = tf.keras.models.load_model('model.keras')
prediction = model.predict(image)
print(prediction)

Training Details

Training Data

From the kaggle zoo challenge the classes one_one(Spheroid) 80%> and one_two(Spiral) 90%> are used.

Furthermore are the image segmented for noice removal

Training Procedure

data_augmentation = tf.keras.Sequential([
  tf.keras.layers.RandomFlip('horizontal'),
  tf.keras.layers.RandomRotation(0.2),
  tf.keras.layers.RandomZoom(0.2),
  tf.keras.layers.RandomContrast(0.2),
  tf.keras.layers.RandomBrightness(0.2),
  tf.keras.layers.GaussianNoise(0.1),
])

AUTOTUNE = tf.data.AUTOTUNE
train_ds = train_ds.cache().shuffle(1000).prefetch(buffer_size=AUTOTUNE)
val_ds = val_ds.cache().prefetch(buffer_size=AUTOTUNE)

model = tf.keras.Sequential([
    data_augmentation,
    tf.keras.layers.Rescaling(1./255, input_shape=(img_height, img_width, 3)),
    tf.keras.layers.Conv2D(32, (3, 3), activation='relu'),
    tf.keras.layers.MaxPooling2D(2, 2),
    tf.keras.layers.Conv2D(64, (3, 3), activation='relu'),
    tf.keras.layers.MaxPooling2D(2, 2),
    tf.keras.layers.Conv2D(128, (3, 3), activation='relu'),
    tf.keras.layers.MaxPooling2D(2, 2),
    tf.keras.layers.Conv2D(128, (3, 3), activation='relu'),
    tf.keras.layers.MaxPooling2D(2, 2),
    tf.keras.layers.Flatten(),
    tf.keras.layers.Dropout(0.5),
    tf.keras.layers.Dense(512, activation='relu'),
    tf.keras.layers.Dense(num_classes, activation='softmax')
])

model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=0.001),
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])

Training Hyperparameters

Speeds, Sizes, Times [optional]

[More Information Needed]

Evaluation

Testing Data, Factors & Metrics

Testing Data

Test data is manual retrieved data from Hubble and James web, see manually manipulated data in the files and their accuracy. Of each a log and linear scaling is used.

Results

          precision    recall  f1-score   support

 one_one       0.96      0.98      0.96      1637
 one_two       0.98      0.93      0.96      1740

accuracy                           0.96      3377

macro avg 0.96 0.96 0.96 3377 weighted avg 0.96 0.96 0.96 3377

Environmental Impact

Hardware Type: M3 Pro
Hours used: 30min