Internal decay is an important quality attribute in chestnuts (Castanea spp.). Worldwide, internal decay is mainly caused by microorganism attack and physiological cell breakdown. It is problematic for the industry, and impacts consumer satisfaction, shelf life, and proper storage. Currently, destructive techniques can be employed to evaluate fresh chestnut internal quality. However, clearly not all produce can be evaluated. In commercial situations, decayed chestnuts are eliminated by their proclivity to float in water. Nonetheless, performance significantly varies between species and throughput, making this floating method unreliable for sorting purposes. Thus, the overall objective of the study is to develop the methods to nondestructively visualize and automatically classify fresh chestnuts, based on their internal quality, using X-ray CT imaging. In this study, medical grade computed tomography (CT) was used to obtain transversal two-dimensional (2D) images from fresh chestnuts (cv. `Colossal' and Chinese seedlings). If the information obtained by the CT scanning of fresh chestnuts is to be used in an industrial setting for in-line sorting, automated interpretation of CT images is essential. For this purpose: (1) Chestnut CT image quality was optimized by studying the combined effect of image acquisition parameters (voltage - 120 kV, current - 170 mA and slice thickness - 2.5 mm) using response surface methodology, (2) effective image visualization techniques to infer fresh chestnut internal quality attributes were established, and (3) an image analysis algorithm for the automatic classification of CT images obtained from 2848 fresh chestnuts, during the harvesting years from 2009 to 2012, was developed and tested.The CT imaging system provided high-resolution and high-contrast images of the internal structure and components of fresh chestnuts. Approximately 50 original CT image slices (stack) were obtained per chestnut, from three different planes (angular orientations) across the longitudinal (Z) (XY-plane-slice), horizontal (YZ-plane-slice) and vertical (XZ-plane-slice) axes. From this image stack, 6 secondary CT images per chestnut sample, including mean and maximum intensity value images for each of the planes were extracted. Thereafter, a total of 1194 grayscale intensity, and textural features were extracted from the 6 secondary CT images per sample. Ultimately, 86, 155 and 126 features were found to be effective in designing a quadratic discriminant analysis classifier with an overall performance accuracy of 85.9 %, 91.2 % and 96.1 % for 5, 3 and 2 classes, respectively. This study provides a powerful tool to accurately visualize and sort chestnuts based on their internal quality, leading to the improved marketability of attractive, safe, high quality chestnuts. Results show that this method is accurate, reliable, and objective and it is applicable to an automatic noninvasive in-line CT sorting system.
Read
