Underwater bubble plumes multi-scale morphological feature extraction and state recognition method.

A large amount of information can be derived from the motion form of underwater bubble plumes. As a result of the inadequate light, mutual adhesion, and absence of a background for the bubble plumes, it is difficult to extract features from the images of bubble plumes, and the recognition accuracy is low. Using a combination of nonsubsampled contourlet transform (NSCT) and quantum neural network (QNN), we present a method for extracting bubble plumes features and recognizing their states. To obtain the multi-scale sub-band images, the underwater bubble plumes image is transformed by NSCT. In the case of low-frequency images, the fuzzy set binarization method is used to extract bright spots, after which morphological features are calculated. The differential box counting (DBC) method is used to calculate the fractal dimension for the high-frequency images, which is used as a directional detail feature. In order to achieve accurate state recognition of the underwater bubble plumes, the quantum gate set convolution neural network (QCSCNN) was designed, taking into account the advantages of the quantum gate and the convolution neural network (CNN). In conclusion, the proposed method is implemented and the experimental results indicate that it achieves the promising and satisfactory results. According to the proposed method, it has been confirmed that it is able to achieve a good effect on convergence speed as well as the recognition accuracy of underwater bubble plumes. [ABSTRACT FROM AUTHOR]