Roof Type Classification with Innovative Machine Learning Approaches
Abstract
In this study, roof type classification was made with a small number of examples for training, in accordance with the one-shot learning approach and using the "Siamese Neural Network" method. The images used for training were artificially produced due to the difficulty of finding roof data. A dataset consisting of real roof pictures was used for the test. The test and training dataset consists of 3 different types: Flat, Gable and Hip. Finally, a CNN-based model and a Siamese Neural Network model were trained with the same datasets and the test results were compared with each other.
Data
- Hip, Flat and Gable roof type of examples are given.
- Cati images are of three types of examples, Flat, Gable, and Hip, and were produced by the authors using Autodesk Maya software
Codes
- one_shot_model.py is the code for create Siamese model.
- one_shot_mydataset.py is the code for read and label the dataset.
- one_shot_train.py is the code for train your dataset and for get results.