OBJECT DETECTION AND CLASSIFICATION IN ASTRONOMICAL IMAGES USING IMAGE PROCESSING AND MACHINE LEARNING
The search to find answers to the deepest questions we have about the universe and the development of observational instrumentation technology has triggered the collection of data onto increasingly large volumes, including billions of astronomical objects and using up to petabytes of memory data siz...
Saved in:
| Main Author: | |
|---|---|
| Format: | Theses |
| Language: | Indonesia |
| Online Access: | https://digilib.itb.ac.id/gdl/view/45925 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Institut Teknologi Bandung |
| Language: | Indonesia |
| Summary: | The search to find answers to the deepest questions we have about the universe and the development of observational instrumentation technology has triggered the collection of data onto increasingly large volumes, including billions of astronomical objects and using up to petabytes of memory data size. So that the detection and classification algorithm is needed automatically and efficiently to be able to process and extract data that can be analyzed further. In this thesis we develop an object detection and classification program using image processing and machine learning used in astronomical imagery. The detection program uses images from the Sloan Digital Sky Survey Data Release Ten (SDSS-DR10) database using the Contrast Limited Adaptive Histogram Equalization (CLAHE), Otsu method, watershed and image moment methods. The program successfully detects objects (such as stars and galaxies) in the SDSS image and can measure some basic parameters of each object. The classification program aims to classify star objects with non-star objects using 14 different machine learning algorithms and is applied to 28,369 objects derived from the data detected in the previous program. Each algorithm is applied to a variety of parameters so that the best parameter will be used when comparing the 14 algorithms. The efficiency of separating stars from non-stars uses a precision function. The program produces 5 best models with precision values of about 98%, namely: random forest, adaboost, gradient boosting, neural network and support vector machines (SVM) with linear kernels.
|
|---|