Fast detection of the early decay in oranges using visible-LED structured-illumination imaging combined with spiral phase transform and feature-based classification model

The early decay of citrus can cause economic and serious food safety issues. The early decayed area has no obvious visual characteristics, making effective detection of this damage very difficult for the citrus industry. This study constructed a new detection system based on visible-light emitting diode (LED) structured-illumination imaging and proposed an effective methodology combined with a spiral phase transform (SPT) algorithm for the early detection of decayed oranges. Each sample obtained three phase-shifting pattern images with phase shifts of −2π/3, 0, and 2π/3 at a spatial frequency of 0.25 cycles/mm. Three strategies (i.e., the conventional three-phase-shifting method, 2-phase SPT, and 1-phase SPT) were used to demodulate the original patterned images to recover the direct component (DC) and amplitude component (AC) images. The partial least squares discriminant analysis (PLS-DA) and least squares support vector machine (LS-SVM) classification models were established based on the texture features of DC, AC, and RT (i.e. the ratio of AC to DC) images. Then, the random frog (RF) algorithm was used to simplify the optimal full-featured model. Finally, the LS-SVM model constructed using 7 texture features from the RT image obtained an average classification accuracy of 95.1% for all tested samples. This study indicates that the proposed structured-illumination imaging technique combined with 2-phase SPT and feature-based classification model can achieve the fast identification of early decayed oranges.