CONVOLUTION NEURAL NETWORK FOR STAR PATTERN RECOGNITION WITH BRIGHTEST STAR AS GUIDE STAR

• Main Author: Wayan Fajar Surya Negara, I
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/74236
• Institution: Institut Teknologi Bandung

The need to determine the attitude and orientation while in "Lost-In-Space (LIS)" is vital for spacecraft navigation. Star tracker, as one of the sensors in spacecraft navigation, will identify star patterns in the field of view (FOV). Star pattern recognition usually will be done by making an algorithm to classify the star image. Convolution neural network (CNN) as one of the best-known deep learning in identifying images, will be used for this research. This process will be more challenging by the fact that the star in the sky has a stochastic probability due to natural conditions. This research presents a method for identifying star patterns with two feature extraction types: multitriangle and spiderweb algorithms. This research will also compare the performance of three different convolution neural networks: VGG16, ResNet, and simple CNN to classify the star based on the feature extraction process. The proposed idea was simulating on the python software and using a simulated star image in the field of view (FOV) with 62.2 × 48.8 degrees dimension and converting that to 3280 × 2464 pixels resolution images finally cropped to 224 × 224 pixels. Simulation results show that the CNN model made is already good, but the image dataset for the simulation needs to be wider to cover more stars.