Background

Lunar samples have its own scientific importance but it differs from others, and its value really depends on the interests of the scientist. The astronaut faces plenty of hours of extravehicular activity(EVA) to evaluate the potential scientific value of the specimen. The samples consists of rocks, rock fragments, core samples, or dust. The zone to obtain the samples will be considered from a landing site. Some of the zone can be very dark and cold, and may have steep sides. The samples that is collected during the mission is preserved by stratification or sometimes included by general mineral content. The samples will soon be evaluated as they are gathered.

Solution

I-Moon has a build machine learning solution where it studies the difference of the lunar samples by capturing images using this camera. I-Moon trains itself as it encounters new type of samples.

Next Steps

• Improve CNN(Convolution Neural Network) model by integrating with YOLO so it enables multi-segmentation for object detection.
• Add sensors to detect the composition of lunar samples that contain metals, dust and halogen.

Impact

Astronaut get to evaluate scientific value of the specimens before and still on the mission as they monitor and differentiate the samples, by that astronauts don't need to spend a long period of time to check the potential scientific value of the samples. Plus NASA don't need to send an bulky and expensive rover to Mars nor the Moon as they can use our solution in more quantity at a lower price and with scalable features.

Resources used from NASA

Lunar Source Book - https://www.lpi.usra.edu/publications/books/lunar_sourcebook/pdf/LunarSourceBook.pdf

Information

Click this link to our website for more info: GLOROC WEBSITE

Source Code

Github repository: https://github.com/Thivedraan/Code-for-NASA-Space-App-Challenge

Short description of our github repo: The interface can be can be run using android studio. The app-debug.apk can be installed to be used directly in an android phone. The Today_model_trainer.py should be trained using datasets of images, then use the model trained in Test_obj.py