A deep multi-resolution approach using learned complex wavelet transform for tuna classification

Tuna fish is commercially important because of its popularity as both canned and raw meat. Tuna account for a significant proportion of the world fishery products. These fishes are separated based on their types in the industry for exporting raw fishes and processed fish products. An automatic system that can classify fishes into different species will ease the work in such industries. The work proposes an automated tuna classification system using textural macro features. The system uses a dual-tree network with signal matched complex wavelet transform. The signal matched wavelet obtained by the lifting scheme is used to develop a complex wavelet. The complex wavelet-based deep architecture is used to extract translation and deformation invariant features. The system performance is evaluated using different classifiers with ten-fold cross-validation. Results show that the proposed system gives an accuracy of 94.58% using cubic support vector machine (SVM) classifier. The proposed system is evaluated using an independent dataset, and the system shows an accuracy of 92.55%, the precision of 92.52%, recall of 92.03%, and F-score of 92.27%. Performance of the system is compared with the state-of-the-art systems using different performance metrics such as accuracy, precision, recall, F-score, and misclassification error.