AI, low-cost hardware

From novel, bespoke sensor design, through to complex data analysis, we find ourselves supporting clients throughout the entire data journey. In this case, we were delivering an Innovate UK project with Newcastle University, to establish the feasibility of quantifying plant nutrient content from RGB images. As a six-month project, we rapidly established the experimental matrix, installed 20 cloud-connected AI-ready cameras, and captured a data set (>10k images) of images with associated metadata of plant nutrition content through conventional lab-based analysis through the chemistry analysis department at Newcastle University, our project partner. 

This data was then used to train a convolutional neural net model, with pre-convolution image processing including background removal, defunct image detection and removal, and dataset augmentation. This model provided the ability to understand the ability of RGB-only image capture for plant nutrient quantification.