Sediment Classification of Small-Size Seabed Acoustic Images Using Convolutional Neural Networks

Abstract

Seabed acoustic images are image data mosaics derived from seafloor acoustic backscattering intensity data, which is related to the type of sediment covering the seabed. Therefore, submarine sediment classification can be realized using seabed acoustic images, and has been studied extensively. Recently, deep learning has also rapidly advanced; in particular, deep convolutional neural networks (CNNs) are now being used to achieve remarkable results in the field of image processing-showing that they are well-suited for image classification tasks. Previous studies have used GoogleNet to classify large-scale side-scan sonar data, with some sediments being well-classified. However, deep learning is data-driven and, theoretically, the greater the depth, the stronger is the learning ability of the feature. It is worth noting that the dataset used for sediment classification can sometimes be small. Hitherto, no related research has analyzed the feasibility and applicability of a CNN classifier for a small-sized seabed acoustic image dataset, so we adopted two different CNN classifier models to conduct the classification experiment in this study. As the results show, the CNN classifier can be applied to the classification of sediments based on a small-sized seabed acoustic image dataset, and the classification performance of shallow CNN was found to be better than that of the deep CNN on existing side-scan sonar data. In particular, the accuracy obtained from the results of several sediment classification experiments using a shallow CNN classifier ranged between 93.4% (Sand Wave) and 87.54% (Reef).INDEX TERMS Deep convolutional neural network, seabed acoustic image, seabed sediment classification.

Underwater Object Classification in Sidescan Sonar Images Using Deep Transfer Learning and Semisynthetic Training Data

“…The first and second rows are airplanes and drowning victims, respectively; and the third and fourth rows are wedge and cylinder mines, respectively. be compared with the method using BOF descriptor [56] on SIFT features [57] and SVM classification, the method using a shallow CNN trained from scratch [40], the gcForest method using Deep Forest [58], the method using deep learning of small datasets [59]. All the methods used for comparison have been demonstrated to be effective on small-scale training data: Before the rise of deep learning, the SIFT-based descriptors usually perform best among various local descriptors [60], and SVM can work well with only small samples [61]; the gcForest method is highly competitive to deep neural networks and can work well even when there are only small-scale training data; the method in [59] transfers all the layers of a pretrained VGG16 network except the last two full connect layers, add a new fully connected layer, and then fine-tunes it to achieve good performance on small datasets.…”

“…For instance, by interleaving multiple convolutional and pooling layers, convolutional neural networks (CNNs), have shown great advantages in object classification [24]- [26], object detection [27], [28], and semantic segmentation [29], [30], sometimes even able to surpass human ability. Recently, CNNs have also been applied to MLOs or marine vessels detection [31]- [37] and seafloor classification [38]- [40], and have been proven to be more effective than traditional methods.…”

A Survey of Underwater Acoustic Data Classification Methods Using Deep Learning for Shoreline Surveillance

“…The last fully connected layers of this model were transferred to a new CNN, which was fine-tuned with a semisynthetic sonar dataset. The results reported in [ 128 ] were compared with more traditional methods for object classification, such as Support Vector Machines, a shallow CNN [ 134 ] and Deep Forests [ 135 ], where the TL-CNN showed better performance and accuracy. However, no comparison was described with respect to state-of-the-art methods for object classification from sonar, such as GNNs [ 62 ] (described at Section 3.2 ).…”

Sediment Classification of Acoustic Backscatter Image Based on Stacked Denoising Autoencoder and Modified Extreme Learning Machine

“…The different types of seabed sediments provide important reference information for scientific research via seabed geological surveys, marine engineering construction, marine space planning, and benthic communities [1][2][3][4]. Traditional snapshot sampling and underwater photography are inefficient and costly for sediment classification [5]. On the contrary, it is feasible to utilize an intensity image of acoustic backscattering to judge the type of underwater sediments present, based on a certain number of samples.…”