Abstract:
A gemstone is a mineral that is produced from geological processes. All gemstones
are either sorted by their colour, nature, or hardness. Gemstones have the potential to
make improvement in the export earnings of Pakistan. Gemstones mines inside
Pakistan can generate 4-5 or more billion dollars. The complexity of the product is
high, audience often suffer loss. It is very difficult to tell difference between real and
fake gemstones. The laboratory identification test is time consuming and expensive.
There is a need of fast and cheap expert gemstone identification system.
The main objective is to create a system that would be able to identify multiple types
of gemstone types, namely {Turquoise, Ruby, and Emerald}. AL-biruni Gemstone is
gemstones identifier based on machine learning. Machine learning is based on the
idea that system can learn from data and forecast prediction on unseen data. Machine
learning includes series of learning techniquses—supervised, unsupervised,
reinforcement learning. The machine learning technique used for design of model is
convolution neural network. The benchmark of the system is to create a system
having accuracy more than 90% with large dataset The model has 12 layers.. The
hyper parameter tuned is epoch, batch size and learning rate. The learning rate used
were 0.001, 0.0001, and 0.00025. The batch sizes were 8, 12, and 16.
The Application is developed as prove of concept named Gemo. Gemo android
Application is used for integration with machine learning model. The application
provides two main features—Authentication and Identification. The authentication
features take image as input and return gemstones type as output. The Accuracy of
the model used in the application for authentication is 95.0%
