Generating Synthetic Sidescan Sonar Snippets Using Transfer-Learning in Generative Adversarial Networks

Steiniger, Yannik; Kraus, Dieter; Meisen, Tobias

doi:10.3390/jmse9030239

Cited by 15 publications

(8 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Considering that methods of transfer learning are unable to achieve optimal results due to the small size of training samples in some special applications, some scholars have focused on the methods of sample generation for enhancing the dataset [8,27,[31][32][33][34]. For instance, in the data enhancement method for sonar images, the pseudo-sample synthesis model takes a conventional optical image and SSS images as inputs to generate a pseudo-SSS image with the content of the optical image but with the characteristics of the SSS image.…”

Section: Transfer Learning From Multi-domainmentioning

confidence: 99%

See 1 more Smart Citation

A Multi-Domain Collaborative Transfer Learning Method with Multi-Scale Repeated Attention Mechanism for Underwater Side-Scan Sonar Image Classification

Cheng

Huo

2022

Remote Sensing

View full text Add to dashboard Cite

Due to the strong speckle noise caused by the seabed reverberation which makes it difficult to extract discriminating and noiseless features of a target, recognition and classification of underwater targets using side-scan sonar (SSS) images is a big challenge. Moreover, unlike classification of optical images which can use a large dataset to train the classifier, classification of SSS images usually has to exploit a very small dataset for training, which may cause classifier overfitting. Compared with traditional feature extraction methods using descriptors—such as Haar, SIFT, and LBP—deep learning-based methods are more powerful in capturing discriminating features. After training on a large optical dataset, e.g., ImageNet, direct fine-tuning method brings improvement to the sonar image classification using a small-size SSS image dataset. However, due to the different statistical characteristics between optical images and sonar images, transfer learning methods—e.g., fine-tuning—lack cross-domain adaptability, and therefore cannot achieve very satisfactory results. In this paper, a multi-domain collaborative transfer learning (MDCTL) method with multi-scale repeated attention mechanism (MSRAM) is proposed for improving the accuracy of underwater sonar image classification. In the MDCTL method, low-level characteristic similarity between SSS images and synthetic aperture radar (SAR) images, and high-level representation similarity between SSS images and optical images are used together to enhance the feature extraction ability of the deep learning model. Using different characteristics of multi-domain data to efficiently capture useful features for the sonar image classification, MDCTL offers a new way for transfer learning. MSRAM is used to effectively combine multi-scale features to make the proposed model pay more attention to the shape details of the target excluding the noise. Experimental results of classification show that, in using multi-domain data sets, the proposed method is more stable with an overall accuracy of 99.21%, bringing an improvement of 4.54% compared with the fine-tuned VGG19. Results given by diverse visualization methods also demonstrate that the method is more powerful in feature representation by using the MDCTL and MSRAM.

show abstract

Section: Transfer Learning From Multi-domainmentioning

confidence: 99%

“…The randomly generated samples with consistent distribution of the training dataset are created by the generative adversarial networks (GAN), which are trained to learn an image-translation from low-complexity ray-traced images to real sonar images [27,34].…”

mentioning

confidence: 99%

A Multi-Domain Collaborative Transfer Learning Method with Multi-Scale Repeated Attention Mechanism for Underwater Side-Scan Sonar Image Classification

Cheng

Huo

2022

Remote Sensing

View full text Add to dashboard Cite

show abstract

“…• Cropping and zooming, which selects a smaller region of the image; • Erasing, the replacement of parts of the image with random noise. Other strategies, that are based on ML, are texture and style transfer [19], and synthetic images created by Generative Adversarial Networks (GAN) [20], [21], [22]. However, in this work, only the geometrical transformations, random cropping, brightness, contrast and sharpness adjustments will be used.…”

Section: Image Data Augmentation Techniquesmentioning

confidence: 99%

Partial Camera Obstruction Detection Using Single Value Image Metrics and Data Augmentation

Oliveira

Niemi

García-Ortiz

et al. 2022

2022 6th International Conference on System Reliability and Safety (ICSRS)

View full text Add to dashboard Cite

To improve the resilience and ensure the dependability of a critical system, the measurements and the derived intelligence provided by the sensors monitoring the system need to be reliable. This is increasingly challenging. As the computer vision methods evolve, the usage of cameras as a part of monitoring solutions has increased, and, consequently, the need for reliable diagnosis strategies for those image-based sensors. This work investigates the suitability of various single-value image metrics, derived from first and second-order statistics, for detecting partial camera obstruction. The presented methodology includes using data augmentation techniques to expand a small dataset of labeled images, and a score-based selection of the best metrics for the target application. The results show that even simple first-order statistics, such as the image histogram skewness, can provide good detection results. The strategy presented could be extended and adapted for the detection of other types of physical anomalies, being particularly useful for integrity assessment in applications with limited computational resources.

show abstract

“…With the development of sample augmentation techniques in the optical imaging field, data augmentation techniques in the underwater acoustics field have also emerged [15][16][17][18][19][20]. Currently, the main methods for Side-scan sonar image augmentation are of two types: one is the image style transfer method represented by GAN (Generative Adversarial Networks) [21][22][23][24][25][26][27][28][29][30][31][32][33], and the other is based on the diffusion model for image generation [34]. For instance, Ye Xiufen [23] used the AdaIN network for style transfer and achieved good results in target detection; Yang Zhiwei [24] adopted an improved DDIM model for data augmentation, successfully enhancing the model's accuracy; Huang Chao [21] utilized the metal style network for data augmentation from geometric and physical perspectives, obtaining a rich set of Side-scan sonar images.…”

Section: Introductionmentioning

confidence: 99%

Sample Augmentation Method for Side-Scan Sonar Underwater Target Images Based on CBL-sinGAN

Peng,

Jin,

Bian

et al. 2024

JMSE

View full text Add to dashboard Cite

The scarcity and difficulty in acquiring Side-scan sonar target images limit the application of deep learning algorithms in Side-scan sonar target detection. At present, there are few amplification methods for Side-scan sonar images, and the amplification image quality is not ideal, which is not suitable for the characteristics of Side-scan sonar images. Addressing the current shortage of sample augmentation methods for Side-scan sonar, this paper proposes a method for augmenting single underwater target images using the CBL-sinGAN network. Firstly, considering the low resolution and monochromatic nature of Side-scan sonar images while balancing training efficiency and image diversity, a sinGAN network is introduced and designed as an eight-layer pyramid structure. Secondly, the Convolutional Block Attention Module (CBAM) is integrated into the network generator to enhance target learning in images while reducing information diffusion. Finally, an L1 loss function is introduced in the network discriminator to ensure training stability and improve the realism of generated images. Experimental results show that the accuracy of shipwreck target detection increased by 4.9% after training with the Side-scan sonar sample dataset augmented by the proposed network. This method effectively retains the style of the images while achieving diversity augmentation of small-sample underwater target images, providing a new approach to improving the construction of underwater target detection models.

show abstract

Generating Synthetic Sidescan Sonar Snippets Using Transfer-Learning in Generative Adversarial Networks

Cited by 15 publications

References 16 publications

A Multi-Domain Collaborative Transfer Learning Method with Multi-Scale Repeated Attention Mechanism for Underwater Side-Scan Sonar Image Classification

A Multi-Domain Collaborative Transfer Learning Method with Multi-Scale Repeated Attention Mechanism for Underwater Side-Scan Sonar Image Classification

Partial Camera Obstruction Detection Using Single Value Image Metrics and Data Augmentation

Sample Augmentation Method for Side-Scan Sonar Underwater Target Images Based on CBL-sinGAN

Contact Info

Product

Resources

About