Machine Learning for Contaminant Detection in X-Ray Spectral Photon Counting

Detection of contaminants in food quality assurance is one of the most important tasks of the food industry. For example, X-ray scanners are routinely used for deboned chicken scanning, but soft cartilaginous bones are often missed by conventional X-ray machines due to lack of attention from the operator and insufficient ability of dual-energy X-ray systems to automatically identify these low-density bone fragments. Choking hazards from soft chicken bones is one of the most common causes of emergency room visits, especially for children. This problem is amenable to be solved using multispectral X-ray imaging due to the difference in calcium content in the bones and meat. This chapter describes one of the first applications of a multienergy spectral X-ray system in nondestructive testing (NDT) food inspection. With a specific interest in identifying food contaminants in X-ray food scanning, for example, chicken bones in tested poultry, this study developed efficient hardware and software solutions to accomplish that objective. The hardware platform used in this study has been described elsewhere while machine learning software used to enhance the detection performance is presented here.