Tuna classification using colour, structure, wavelet, and fuzzy based textural features

Tuna is one among the popular commercial fishes and forms a major proportion of world fishery products. Separation of these fishes into different species is important to export raw fish and processed fish products. The work proposes an automatic system to separate tuna species using image processing techniques. Images are cropped to reduce the background effect and are enhanced using adaptive histogram equalization. Statistical features from colour histogram, structural features from histogram of oriented gradients, textural features from fuzzy based local binary pattern, and wavelet transform based statistical features are used for the classification. Size of the feature set is decreased by an ensemble of dimensionality reduction techniques. Performance of the system is evaluated with both non-reduced and reduced feature set using different classifiers for ten-fold cross validation. Results show that extreme learning machine classifier using Mortlet wavelet kernel gives high classification accuracy of 92.95% and 90.85% with non-reduced and reduced feature vector respectively.