Optical property mapping and early-bruise identification in apples using spatial frequency domain imaging

Apple bruises are a common defect that diminishes fruits' economic value. Therefore, an effective method for detecting fruit bruises is crucial. Nondestructive testing methods, such as near-infrared and hyperspectral imaging typically use tissue absorption properties, disregarding scattering as noise even if fruit tissues are highly scattering media. This study created a spatial frequency domain imaging system to identify early apple bruises by focusing on scattering properties. The optical parameters of bruised and nonbruised apple tissues were determined using a new inversion method proposed in this paper. Comparative analysis shows that the alternating current (AC) component is more effective in detecting early bruises than the direct current (DC) component. The ratio of the AC to DC components in structured-light images helped in identifying bruises. Suspicious points are identified using Difference of Gaussians images and Hough circle detection, where these points correspond to potential locations of bruises. Subsequently, subimages were extracted and a classification algorithm was used to evaluate the presence of bruises within those subimages. The classification accuracies of decision trees, linear discriminant analysis, K-nearest neighbors, and support vector machines (SVM) are 94.56 %, 98.12 %, 97.75%, and 99.81 %, respectively. Ultimately, SVM combined with a localization algorithm achieved a classification accuracy of 98.78 % for apples and accurately identified the locations of all bruises. The results demonstrate that the proposed method achieves remarkable detection accuracy, providing a new approach for detecting apple bruises.