COMPARISON STUDY OF GAMMA RAY SPECTRUM IDENTIFICATION MODELS FOR BA133, CO60, CS137, AND NA22 USING YOU ONLY LOOK ONCE (YOLO) AND 1D CONVOLUTIONAL NEURAL NETWORK (1D CNN) METHODS

• Main Author: Michele Simbolon, Elizabeth
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/83912
• Institution: Institut Teknologi Bandung

With advances in computational technology, the identification of Gamma Ray spectra has become faster and more accurate, allowing for efficient isotope analysis. This study aims to develop and evaluate deep learning models to identify Gamma Ray spectra from isotopes Ba133, Co60, Cs137, and Na22, as well as combined spectra of these isotopes. Gamma Ray spectra are important in nuclear physics and medical applications because each isotope has a unique spectrum. The research methodology includes literature review, data collection of Gamma Ray spectra, data preprocessing, and data augmentation. The processed data is then used to train and test YOLO and 1D-CNN models. The YOLO model is tested with variations in spectrum image and color spectrum inputs, while the 1D-CNN model is used to analyze sequential data. The results show that the 1D-CNN model achieves the highest accuracy with an average of 96.21%. The YOLO model’s second variation (color spectrum) achieves an accuracy of 82.69%, while the first variation (spectrum image) achieves 54.17%. The 1D-CNN model performs best in identifying combined spectra. The combination of YOLO and 1D-CNN models improves the efficiency and accuracy of Gamma Ray spectrum identification. Further research is recommended to increase the amount of training data and to use more varied data augmentation techniques.