Surface defect identification of Citrus based on KF-2D-Renyi and ABC-SVM

In allusion to the problems of citrus surface defect identification such as blurred edges, unclear images, more interference and difficulty in defect identification, surface defect identification of citrus based on KF-2D-Renyi and ABC-SVM was proposed in this paper. First, the method based on the dark channel prior (DCP) was used to defog the citrus images collected. Then, the firefly algorithm based on Kent chaos was used to optimize two-dimensional Renyi entropy threshold segmentation algorithm (2D-Renyi). The citrus surface defects were segmented, and the image features were extracted. Finally, the image feature vectors were input into the ABC-SVM classifier to determine the citrus defect types. We selected 8 kinds of citrus surface defects to carry on the experiment. In testing the segmentation algorithms, compared with the traditional threshold segmentation algorithms, the KF-2D-Renyi segmentation algorithm has a great improvement. The recognition rates for the defects whose features are obvious such as Sooty mould and Anthracnose could reach 100%. The recognition rates for the defects which are difficult to identify such as Thrips scar, Oleocellosis and Scale injury reached 95.18%, 96.37% and 98.43% respectively. In testing the classification algorithms, compared with the standard SVM classifier, the PSO-SVM classifier and the neural network classifiers, the average recognition rate of the ABC-SVM classifier reached 98.45%. The experimental results show that the method in this paper can effectively detect and classify citrus surface defects.