Author: Ibrahim Oulad Amar
Final degree project which consists of creating a meteor classification system using Deep Learning techniques.
The classifier uses 256x256 images as input. The images shall be the MAXPIXEL ones in the FTP format. In this case I used the data provided by the University of Western Ontario. I am thankful to researcher Denis Vida for his help in obtaining these data.
The data was split in two sets, training (85%) and validation (15%). The model is a CNN + MaxPool + BatchNormalization (total 12 layers) along with 3 fully connected layers. The total number of parameters is 49,449, of which 512 are not-trainable. The model performance metrics are:
- Model Precision: 0.931 (93.1%)
- Model Recall: 0.941 (94.1%)
- Model F1 Score: 0.936
The model is the one defined in the file final_model_weights.h5 in the folder
meteor_sort/results/weights/.
Finally, you can find the thesis document in the root of the project with the name TFG_Oulad_Amar_Ibrahim.pdf or read it directly from here.
If you have any suggestions or questions don't hesitate to reach me at my email: ibrahim.ouladamar@gmail.com