Smart spectral vision system

Abstract

Spectral imaging collects images in different bands or sub-bands of the electromagnetic spectrum to obtain extra information that human vision cannot capture. An image recognition system uses shapes for individual identification. The current image recognition system operates in the visible range of the electromagnetic spectrum. Current image recognition systems operate in the visible spectrum from 350nm to 740nm wavelengths. Nowadays, image recognition technology is easily seen as applicable in daily lives, from face recognition for security purposes to even tracking humans in public via CCTVs. Current image recognition technology mainly uses 2D images or 3D images to feed as data to train the neural network.

This project uses a cheap and portable camera module (ESP32-CAMERA) which will then be modified as a spectrometer to obtain spectral images through Arduino IDE services. Spectral images will be collected and analysed to obtain the intensity of different wavelengths.

This project is successful in programming the ESP32-CAMERA to obtain spectral images and measuring the intensity of the wavelength. This paper investigates the feasibility of using deep learning models in neural networks to use spectral images for image recognition.