Determination of color quality of rambutan based on computer vision

In view of drawbacks of rambutan grade identification, which still relies on human sensory evaluation, the determination of color quality of rambutan based on computer vision was studied. The images of ranbutan were obtained by CCD camera. Image processing algorithm of OSTU and area labeling were applied to 48 rambutan images in order to remove the images of background and stem, and the texture features of color and lustre were extracted. A multi-grade support vector machine (DAGSVM) was set up, and color and lustre features were the inputs of DAGSVM. The results show that the accuracies of four color grades rambutan recognized by SVM model were 94%, 88%, 89%, 95% respectively with good stability. The identification speed and ability of this model are much superior to algorithm neural network.