Nondestructive internal disorders detection of ‘Braeburn’ apple fruit by X-ray dark-field imaging and machine learning

Abstract

'Braeburn' apples are susceptible to internal browning disorders when stored under controlled atmosphere (CA) conditions with unfavorable gas compositions. The progression of CA-related disorders in apple tissues is dynamic, noting a decrease in porosity during early storage due to cellular breakdown and pore flooding, and an increase in porosity in later stages due to structural collapse and cavity formation. Utilizing grating-based X-ray dark-field radiography, which leverages X-ray small-angle scattering to detect microstructural changes below the pixel scale, this study assesses the technique's efficacy in identifying internal disorders in 'Braeburn' apples at both early and later stages. A machine learning approach was applied to compare the diagnostic capabilities of dark-field imaging with those of X-ray absorption radiography at identical image resolutions. Results indicate that for early-stage disordered fruit detection, X-ray dark field radiography is 10 % more accurate than absorption radiography, regardless of the machine learning classifiers that were applied. In the later stage of browning, dark-field imaging performs similarly to absorption imaging. High-resolution micro-computed tomography scans suggested that the distinct detection performance of dark-field imaging may be attributed to the more pronounced microstructural differences between healthy and early-stage defective tissues than those between healthy and later-stage defective tissues. The insights from this work will guide the application of X-ray dark-field systems in fruit quality assurance, particularly in detecting internal disorders.