< Back to previous page

Project

Cellular segmentation of X-ray micro-CT data of fruits and vegetables using deep learning

Knowledge of spatial and dynamic changes of internal food microstructure improves understanding of food quality and serves a basis for developing new ways of producing, storing and processing of foods. Considering that food microstructure is essentially a complex three-dimensional (3D) multiscale property demands for further advancements in the area of 3D visualisation technology. A highly valuable imaging technique to study food microstructure is Computed Tomography (CT). CT is a well-known, nondestructive 3D inspection technique for producing cross-sectional images of an object based on the differences in X-ray absorption by the type and density of the constituents of the food. With novel phase contrast CT, images can be acquired with unprecedented contrast far surpassing conventional CT contrast exploiting the fact that when X-rays pass through a material, a phase shift occurs across the X-ray wavefront, depending on the electron density and the spatial frequency of the material features. The aim of this PhD project is to explore the potential of phase constrast for food applications, and develop advanced image processing algorithms to quantify relevant 2D and 3D microstructural features of foods. In a first step, artificial (foams, prints) and real food microstructures (with a focus on fruit and vegetables) will be prepared and characterized with conventional and phase contrast imaging techniques as to create a reference database, also exploiting already available datasets and protocols. Dedicated phase contrast imaging protocols will be developed and optimized, and applied to the food systems. In a second step, image analysis algorithms are developed for the multimodal processing (combining absorption and refraction signals) of phase contrast images. Multi-feature and multimodal Markov random fields will be considered, as well as deep learning methods using generative adversarial networks. Data augmentation using model-based simulation will be exploited to enlarge training datasets. Finally, successful phase contrast imaging is used to create enhanced 3D food microstructure models for structural engineering applications.

Date:1 Oct 2020 →  Today
Keywords:Phase contrast X-ray imaging, Food microstructure
Disciplines:Image processing, Agrofood mechatronics, Food physics
Project type:PhD project