For the purpose of improving the accuracy of underwater acoustic target recognition with only a small number of labeled data, we proposed a novel recognition method, including 4 steps: pre-processing, pre-training, fine-tuning and recognition. The 4 steps can be explained as follows: (1) Pre-processing with Resonance-based Sparsity Signal Decomposition (RSSD): RSSD was firstly utilized to extract high-resonance components from ship-radiated noise. The high-resonance components contain the major information for target recognition. (2) Pre-training with unsupervised feature-extraction: we proposed a one-dimensional convolution autoencoder-decoder model and then we pre-trained the model to extract features from the high-resonance components. (3) Fine-tuning with supervised feature-separation: a supervised feature-separation algorithm was proposed to fine-tune the model and separate the extracted features. (4) Recognition: classifiers were trained to recognize the separated features and complete the recognition mission. The unsupervised pre-training autoencoder-decoder can make good use of a large number of unlabeled data, so that only a small number of labeled data are required in the following supervised fine-tuning and recognition, which is quite effective when it is difficult to collect enough labeled data. The recognition experiments were all conducted on ship-radiated noise data recorded using a sensory hydrophone. By combining the 4 steps above, the proposed recognition method can achieve recognition accuracy of 93.28%, which sufficiently surpasses other traditional state-of-art feature-extraction methods.

Marine animal studies are of great importance to human beings and instrumental to many research areas. How to identify such animals through image processing is a challenging task that leads to marine animal segmentation (MAS). Although deep neural networks have been widely applied for object segmentation, few of them consider the complex imaging condition in the water and the camouflage property of marine animals. To this end, a robust deep marine animal segmentation network is proposed in this article. Specifically, we design a new data augmentation strategy to randomly change the degradation and camouflage attributes of the original objects. With the augmentations, a fusion-based deep neural network constructed in a Siamese manner is trained to learn the shared semantic representations. Moreover, we construct a new large-scale real-world MAS data set for conducting extensive experiments. It consists of over 3000 images with various underwater scenes and objects. Each image is annotated with an object-level mask and assigned to a category. Extensive experimental results show that our method significantly outperforms 12 state-of-the-art methods both qualitatively and quantitatively.

Ship recognition based on ship-radiated noise is one of the most important and challenging subjects in underwater acoustic signal processing. The recognition methods for ship-radiated noise recognition include traditional methods and deep learning (DL) methods. Developing from the DL methods and inspired by audio–video speech recognition (AVSR), the paper further introduces multimodal deep learning (multimodal-DL) methods for the recognition of ship-radiated noise. In this paper, ship-radiated noise (acoustics modality) and visual observation of the ships (visual modality) are two different modalities that the multimodal-DL methods model on. The paper specially designs a multimodal-DL framework, the multimodal convolutional neural networks (multimodal-CNNs) for the recognition of ship-radiated noise. Then the paper proposes a strategy based on canonical correlation analysis (CCA-based strategy) to build a joint representation and recognition on the two different single-modality (acoustics modality and visual modality). The multimodal-CNNs and the CCA-based strategy are tested on real ship-radiated noise data recorded. Experimental results show that, using the CCA-based strategy, strong-discriminative information can be built from weak-discriminative information provided from a single-modality. Experimental results also show that as long as any one of the single-modalities can provide information for the recognition, the multimodal-DL methods can have a much better multiclass recognition performance than the DL methods. The paper also discusses the advantages and superiorities of the multimodal-Dl methods over the traditional methods for ship-radiated noise recognition.

In turbid water, the attenuation and scattering of light caused by scatterers make underwater optical images degraded, blurred, and contrast reduced, limiting the extraction and analysis of information from images. To address such problems, a turbid underwater image enhancement method based on parameter-tuned stochastic resonance (SR) is proposed in this article. First, an SR algorithm framework for underwater image enhancement is constructed, including the dimensionality reduction and normalization of input images, the solution and parameter optimization of the SR system, the dimensionality upgrading of output images, etc. This framework can apply the SR's ability to enhance weak signals to the enhancement of turbid underwater images. Second, to measure the performance of the system, a synthetic turbid underwater image data set (UWCHIC) is constructed using the underwater imaging model and an image set with simulated scatterers. Based on this data set, the relationship between various image quality evaluation metrics and system parameters is analyzed, and then the suitable no-reference (NR) metrics for system performance evaluation are selected and an adaptive parameter tuning strategy of the SR system is proposed to guide the image enhancement. Lastly, the proposed method is evaluated on the UWCHIC, a dataset to evaluate underwater image restoration methods (TURBID), marine underwater environment database (MUED), and underwater image enhancement benchmark (UIEB) data sets and the turbid underwater images captured from natural waters. Different experimental evaluations demonstrated that the proposed method not only effectively enhances the visual quality of turbid underwater images but also improves the performance of downstream vision tasks.

Computer vision plays an important role in scientific research, resource exploration, and other underwater applications. However, it suffers from the severe color distortion, which is caused by the scattering and absorption of light in the water. In this article, an underwater image color restoration network (UICRN) is proposed to obtain the real color of the image by estimating the main parameters of the underwater imaging model. First, an encoder neural network is applied to extract features from the input underwater image. Second, three independent decoders are used to estimate the direct light transmission map, backscattered light transmission map, and veiling light. Third, the loss functions and the training strategy are designed to improve the performance of restoration. As we know, the learning-based method would require a paired data set for training. An underwater image generation method is also proposed in this article to obtain the data set consisting of color-distorted images and corresponding ground truth. The method combines the inherent optical properties and apparent optical properties with structure information to generate the paired data set. More than 20 000 pairs of underwater images are generated based on the method. Finally, the UICRN method is quantitatively evaluated through various experiments, such as color chart testing in the South China Sea and natural underwater image evaluation. It demonstrates that the UICRN method is competitive with previous state-of-the-art methods in color restoration and robustness.