Recognition of Wizened Red Jujube Based on Ununiformities of Gradient Distribution

Abstract:

A method of texture feature extraction based on the a* components of L* a* b* color space model was proposed to recognize wizened red jujube from the mix of wizened red jujube and full red jujube. Firstly, grayscale, R component from RGB color space model, a* component from L* a* b* color space model were selected and the gradient values with different gradient operators were calculated respectively. Then, abnormal gradient values were removed by morphological operator and logical operation. Finally, after gradient normalization, the gradient distributions exhibited different behavior for the wizened red jujube and full red jujube. Thus, a texture feature extraction method based on this property was designed, and the recognition of wizened red jujube was achieved. In the experiments, a 12channel fruits and vegetables sorting machine was used to collect 240 images of full red jujube and 202images of wizened red jujube. The experimental results showed that utilizing a* component from L* a* b* color space model can achieve better recognition rate than utilizing grayscale or R component from RGB color space model; simple gradient operator was more suitable than Sobel gradient operator in texture extraction for wizened red jujube recognition; using the normalized gradient histogram as a texture feature representation and the calculated ununiformities of gradient distribution as discriminant features, the recognition rate was over 99% and the false positive rate was 0.83%.