Detection of Potato Slight Bruise Based on Hyperspectral Image and Fruit Fly Optimization Algorithm

Abstract:

Potato is an indispensable food crop for the people in the world. As a kind of light injury on the surface of potato，slight bruise of potato cannot be accurately tested when potato placed in random orientation. This paper proposed a method by combining hyperspectral image based on Vshaped plane mirror with fruit fly optimization algorithm (FOA) to identify slight bruise of potato randomly placed. In this study, hyperspectral imaging system was built based on Vshaped plane mirror and 322 potato samples were bought as the research subjects. To meet with the practical production, within half an hour after bruise occurred, potatoes were placed in three positions: the damage part facing to camera, side to camera, and back to camera. Then hyperspectral images of all potatoes were collected including reflection image F1 in mirror 1, image F2 directly obtained by camera and reflection image F3 in mirror 2. Average spectrums from these three images were spliced into attribute matrix of sample. Support vector classifier(SVC) model was established in full bands after utilizing standard normal variate(SNV) and the recognition accuracy of prediction set was only 8411%. Variable selection was processed by ant colony optimization(ACO). Nine spectral variables (762nm, 879nm in F1; 711nm, 957nm, 1020nm in F2; 510nm, 746nm, 1.000nm, 1.007nm in F3 )were selected and the recognition rate reached 95.32%. FOA, genetic algorithm(GA) and grid search were respectively applied to search the best penalty parameter c and kernel function parameter g. By comparing results of those models, FOA obtained optimal parameters(c=11.0763，g=9.2625). FOA-SVC was proved to be the best model and the training set and prediction set recognition accuracy both reached 100%. The results show that the combination of hyperspectral image based on Vshaped plane mirror with FOA-SVC could accurately detect the slight bruise of potato.