Recently, learning-based algorithms have shown impressive performance in underwater image enhancement. Most of them resort to training on synthetic data and obtain outstanding performance. However, these deep methods ignore the significant domain gap between real (i.e., inter-domain gap), thus models trained often fail generalize well real-world scenarios. Moreover, complex changeable environment also causes a great distribution among itself intra-domain gap). almost no research focuses this...

Forward-looking sonar (FLS) imagery system plays a significant role in oceanic object recognition and detection since it can overcome the limitation of lighting conditions reflect real situation underwater environment. However, algorithms for FLS images remain challenging two main reasons: 1) noise caused by coherent characteristic scattering phenomenon impairs detector capture target information 2) scene prior based on uneven scale distribution is generally neglected, which leads to...

Since underwater images are seriously degraded due to the attenuation of light, artificial light (AL) is often used assist photography in underwater. However, normal imaging process changed by AL. It observed that AL source typically alters condition a large extent, resulting non-uniform illumination images. In addition, color distortion area affected little because close object suffers attenuation. most existing image enhancement algorithms ignore this phenomenon. their results, areas tend...

Underwater images suffer from severe color casts, low contrast and blurriness, which are caused by scattering absorption when light propagates through water. However, existing deep learning methods treat the restoration process as a whole do not fully consider underwater physical distortion process. Thus, they cannot adequately tackle both scattering, leading to poor results. To address this problem, we propose novel two-stage network for image (UIR), divides into two parts viz. horizontal...

With the development of convolutional neural networks (CNNs) in recent years, network structure has become more and complex varied, achieved very good results pattern recognition, image classification, object detection tracking. For CNNs used for addition to structure, researches focus on improvement loss function, so as enlarge inter-class feature differences, reduce intra-class variations soon possible. Besides traditional Softmax, typical functions include L-Softmax, AM-Softmax, ArcFace,...

Underwater enhanced images (UEIs) are affected by not only the color cast and haze effect due to light attenuation scattering, but also over-enhancement texture distortion caused enhancement algorithms. However, existing underwater image quality assessment (UIQA) methods mainly focus on inherent optical imaging, ignore widespread artificial distortion, which leads poor performance in evaluating UEIs. In this paper, a novel mapping-based representation is proposed. We divide into different...

The urban environment has a great impact on the wellbeing of citizens and it is significance to understand how perceive evaluate places in large scale region provide scientific evidence support human-centered planning with better environment. Existing studies for assessing perception have primarily relied low efficiency methods, which also result evaluation accuracy. Furthermore, there lacks sophisticated understanding correlate built other socio-economic data, limits their applications...

Underwater images suffer from severe color casts, low contrast, and blurriness, which greatly degrade the visibility fidelity of underwater images. Recently, numerous image enhancement (UIE) algorithms have been proposed. Existing synthetic datasets-based deep learning methods employ datasets to train UIE models. However, there is a gap between real images, leading poor generalization methods. Besides, existing largely focus on minimizing mean squared reconstruction error results...

Underwater images generally suffer from color cast and haze effects due to light attenuation scattering, which leads image quality degradation poor recognition of content by autonomous machines. Most the existing enhancement algorithms try remove these distortions underwater but do not perform perfectly. Moreover, there is no evaluation metric that can accurately measure enhanced results. Thus, evaluating one urgent problems be solved in imaging research. In this article, a prior-based...

Reconstructing a three-dimensional (3D) structure from single two-dimensional training image (TI) is challenging issue. Multiple-point statistics (MPS) an effective method to solve this problem. However, in the traditional MPS method, errors occur while statistical features of reconstruction, such as porosity, connectivity, and structural properties, deviate those TI. Due reconstruction mechanism that voxel being reconstructed dependent on voxel, it may cause error accumulation during...

Image acquisition and reconstruction play an important role in underwater detections explorations. However, the limited acoustic communication channels narrow bandwidth resources will have a great impact on performance of traditional data methods, resulting loss details blur reconstructed images. Compressive sensing theory (CS) which can reconstruct images from fewer measurement than that required by Nyquist sampling law has been proved to good effect image reconstruction. Nevertheless,...

To analyze the anthropogenic activities on water-environment conditions, sampling campaigns and isotope measurement for precipitation, river water groundwater were conducted in Jinjiang River System, Chengdu, China one year. The results show that runoff is main factor affecting isotopic composition natural-state rivers, while rivers affected by human activities, varies greatly due to influence of impermeable surface municipal pipe networks evaporation mixing. warm-and-humid moisture source...

Underwater images are often affected by color cast and blurring, which degrade the performance of underwater machine vision tasks. While existing image enhancement (UIE) methods have been proposed to improve quality for human perception, their effectiveness in enhancing is limited. In this article, a novel unsupervised UIE framework based on disentangled representation (DR) proposed, designed Specifically, disentangles into two parts latent space according whether they beneficial tasks:...