UC Berkeley

Void fraction and computational tomographic (CT) scans have a wide range of real world applications. X-ray imaging and computed tomography can benefit the agricultural industry by enabling non-destructive inspection and characterization of fresh produce and processed food products. The overall goal is to examine the feasibility of validating the accuracy of simulated training data, which will eventually be used in a broader artificial intelligence (AI) system, by using a ME3 detector. An ME3 detector is a multi-energy 128 threshold(s) cadmium telluride (CdTe) detector. The validation process involves obtaining data from observed experiments and generating simulated data. The observed and simulated data are then each separately inputted into a void fraction equation. The results of the void fraction equation provide an understanding of the accuracy with which the detector can characterize the density between different objects. The observed and simulated data is also compared to each other, without being processed through the void fraction equation, to generate a graph that compares photon count to energy intensity in kilovolts (keV). This graph reveals potential variables that may contribute to the difference between observed data and simulated data, such as Compton Scattering and Poisson noise.

The void fraction equation results and the comparative graph of photon count to energy intensity then serves as the foundational understanding of the CT scans that were taken of a six phase phantom disk, of a phantom disk composed of three types of metal materials, and of a solid water phantom disk. Each material used in the phantom disks were selected to hone the accuracy of different aspects of the simulated data. The six phase phantom disk, which is composed of six different liquids, is used to examine the accuracy with which the ME3 detector can differentiate between the multiphase composition of an object. The phantom disk composed of three types of metal materials test the limits of the resolution of the ME3 detector when an object is close in size to the detector's pixel size. The solid water phantom disk is used as a control to the fluid water that comprises one of the six phases of the six phase phantom disk. In pushing the limits of various components of an ME3 detector, a better understanding is gained of the factors that can affect simulated training data, which assumes ideal conditions. This in turn provides the tools needed to validate the accuracy of the simulated training data.

Ultimately, the understandings gained through rigorous experiments, use of void fraction equation, and analyzing CT scans of the different phantoms serve to sharpen the accuracy of the simulated training data. The training data will then be used to create an AI model, which will ideally one day automate the discretization of foreign objects within the agricultural food supply chain.