DeepSpectra

Warning: This experiment is not a complete project. Because of lack of data both training and validation accuracy is less.

Introduction

I used to work with scientific spectral data and thought of applying machine learning in these data. In this project I have taken powder xrd spectral data to classify it into seven crystal systems. The powder xrd spectrum has two main sequence a 2theta angle in x axis and intensity in y axis. I converted this sequence data into a tensor data to pass it as an image format in a convolution model. This classification is similar to other CNN image classification except instead of passing colour channels as depth here we are passing feature vectors. Number of matrix in the created tensor represents the number of features of the sequence. The possible application is classifiction of uv,ir,raman spectra or protein/nucleotide analysis. . The idea is to find spatial or temporal features (if any) from the sequence data and then classify it into different classes.

DeepSpectra Github Repo