A neural network approach to ship track-keeping control

This paper presents an on-line trained neural net work controller for ship track-keeping problems. Following a brief review of the ship track-keeping control development since the 1980's, an analysis of various existing backpropagation-based neural controllers is provided. We then propose a single-input multioutput (SIMO) neural control strategy for situations where the exact mathematical dynamics of the ship are not available. The aim of this study is to build an autonomous neural controller which uses rudder to regulate both the tracking error and heading error. During the whole control process, the proposed SIMO neural controller adapts itself on-line from a direct evaluation of the control accuracy, and hence the need for a “teacher” or an off-line training process can be removed. With a relatively modest amount of quantitative knowledge of the ship behavior, the design philosophy enables real time control of a nonlinear ship model under random wind disturbances and measurement noise. Three different track-keeping tasks have been simulated to demonstrate the effectiveness of the training method and the robust performance of the proposed neural control strategy     

Sea-floor classification using multibeam echo-sounding angular backscatter data: a real-time approach employing hybrid neural network architecture

The presently studied numerical model, e.g., composite roughness, is successful for the purpose of seafloor classification employing processed multibeam angular backscatter data from manganese-nodule-bearing locations of the Central Indian Ocean Basin. Hybrid artificial neural network (ANN) architecture, comprised of the self-organizing feature map and learning vector quantization (LVQ), has been implemented as an alternative technique for sea-floor roughness classification, giving comparative results with the aforesaid numerical model for processed multibeam angular backscatter data. However, the composite-roughness model approach is protracted due to the inherent need for processed data including system-gain corrections. In order to establish that tedious processing of raw backscatter values is unessential for efficient classification, hybrid ANN architecture has been attempted here due to its nonparametric approach. In this technical communication, successful employment of LVQ algorithm for unprocessed (raw) multibeam backscatter data indicates true real-time classification application.     

A new approach to the structural features of the Aegean Sea: Cellular neural network

In this study, structural features in the Aegean Sea were investigated by application of Cellular Neural Network (CNN) and Cross-Correlation methods to the gravity anomaly map. CNN is a stochastic image processing technique, which is based on template optimization using neighbourhood relationships of pixels, and probabilistic properties of two-Dimensional (2-D) input data. The performance of CNN can be evaluated by various interesting real applications in geophysics such as edge detection, data enhancement and separation of regional/residual potential anomaly maps. In this study, CNN is used in edge detection of geological bodies closer to the surface, which are masked by other structures with various depths and dimensions. CNN was first tested for (prismatic) synthetic examples and satisfactory results were obtained. Subsequently, CNN/Cross-Correlation maps and bathymetric features were evaluated together to obtain a new structural map for most of the Aegean Sea. In our structural map, the locations of the faults and basins are generally in accordance with the previous maps from restricted areas based on seismic data. In the southern and southeastern parts of the Aegean Sea, E–W trending faults cut NE–SW trending basins and faults, similar to on-shore Western Anatolia. Also, in the western, central and northern parts of the Aegean Sea, all of these structures are truncated by NE-trending faults.     

A controlling factor approach to estuary classification

A new approach to the classification of estuaries is described. The estuary environment classification (EEC) is based on a hierarchical view of the abiotic components that comprise the environments of estuaries. The EEC postulates that climate, oceanic, riverine and catchment factors ‘control’ a hierarchy of processes and broadly determine the physical and biological characteristics of estuaries. The classification differentiates estuaries at four levels of detail. Level 1 differentiates global scale variation based on differences in climatic and oceanic processes, which are discriminated by the factors: latitude, oceanic basins and large landmasses. Level 2 differentiates variation in estuary hydrodynamic processes, which are discriminated by estuary basin morphometry, river and oceanic forcing. Level 3 differentiates variation among estuaries that are due to catchment processes, which are discriminated by catchment geology and catchment land cover. The approach has been applied to all the estuaries in New Zealand using existing data sources. Estuaries were assigned class membership at each level of the classification by applying criteria in the form of decision rules to the database of assignment characteristics. GIS was then used to map the estuaries with classes being defined by colour at any level of the classification. The resulting map provides a multi-scale spatial framework that is suitable for many environmental or conservation management applications.     

An automatic approach to the detection and extraction of mine features in sidescan sonar

Mine detection and classification using high-resolution sidescan sonar is a critical technology for mine counter measures (MCM). As opposed to the majority of techniques which require large training data sets, this paper presents unsupervised models for both the detection and the shadow extraction phases of an automated classification system. The detection phase is carried out using an unsupervised Markov random field (MRF) model where the required model parameters are estimated from the original image. Using a priori spatial information on the physical size and geometric signature of mines in sidescan sonar, a detection-orientated MRF model is developed which directly segments the image into regions of shadow, seabottom-reverberation, and object-highlight. After detection, features are extracted so that the object can be classified. A novel co-operating statistical snake (CSS) model is presented which extracts the highlight and shadow of the object. The CSS model again utilizes available a priori information on the spatial relationship between the highlight and shadow, allowing accurate segmentation of the object's shadow to be achieved.     

基于神经网络方法的C波段和Ku波段统一地球物理模型

通过地球物理模型建立后向散射系数与海面风矢量的关系,可将散射计从不同方位角测得的风矢量单元后向散射系数反演得到风矢量,因此地球物理模型在风速反演中起着至关重要的作用。使用神经网络方法,利用C波段经验模型CMOD4和Ku波段经验模型QSCAT—1仿真数据建立了形式统一的C波段和Ku波段地球物理模型。新模型将电磁波频率作为模型的参数之一,使新模型不再局限于特定的传感器,并使C波段与Ku波段具有统一的形式。分析表明,由新模型建立的后向散射系数与海面风矢量的关系同经验模型具有很好的可比性。利用新模型反演的风速与CMOD4和QSCAT—1模型反演的风速具有很好的一致性,说明新模型在具有统一简洁形式的同时也兼有与经验统计模型相同的有效性。     

A unified C-band and Ku-band geophysical model function determined by neural network approach

The geophysical model function （GMF） describes the relationship between backscattering and sea surface wind, so that wind vec- tors can be retrieved from backscattering measurement. The GMF plays an important role in ocean wind vector retrievals, its performance will directly influence the accuracy of the retrieved wind vector. Neural network （NN） approach is used to develop a unified GMF for C-band and Ku-band （NN-GMF）. Empirical GMF CMOIM and QSCAT-1 are used to generate the simulated training data-set, and Gaussian noise at a signal noise ratio of 30 dB is added to the data-set to simulate the noise in the backscat- tering measurement. The NN-GMF employs radio frequency as an additional parameter, so it can be applied for both C-band and Ku-band. Analyses show that the %predicted by the NN-GMF is comparable with the σpredicted by CMOIM and QSCAT-1. Also the wind vectors retrieved from the NN-GMF and empirical GMF CMOIM and QSCAT-1 are comparable, indicating that the NN-GMF is as effective as the empirical GMF, and has the advantages of the universal form.     

A neural approach to the assignment algorithm for multiple-targettracking

A neural network is presented for performing data association for multiple-target tracking on an optimal assignment basis, i.e., the sum of likelihood functions of measurement-to-track file associations is optimized. The likelihoods are shown to be derivable from a Kalman filter, which updates and maintains the track files from the measurements assigned by the neural network. Not only are measurements assigned to track files on an optimal basis, but undetected targets and unassigned measurements are identified also. A multiple-target tracking system utilizing the neural network, in conjunction with Kalman filtering, can also automatically delete and initiate track files. The solution to the data association problem, and therefore the design of the neural network, is based on the minimization of a properly defined energy function. Computer simulations indicate the ability of the neural network to converge quickly to the optimal hypothesis under various conditions, provided that the ambiguity in the scenario is not extreme. The computational complexity involved is moderate     

A neural network based hybrid system for detection,characterization, and classification of short-duration oceanic signals

A comprehensive classifier system is presented for short-duration oceanic signals obtained from passive sonar, which exhibit variability in both temporal and spectral characteristics even in signals obtained from the same source. Wavelet-based feature extractors are shown to be superior to the more commonly used autoregressive coefficients and power spectral coefficients for describing these signals. A variety of static neural network classifiers are evaluated and are shown to compare favorably with traditional statistical techniques for signal classification. The focus is on those networks that are able to time-out irrelevant input features and are less susceptible to noisy inputs, and two new neural-network-based classifiers are introduced. Methods for combining the outputs of several classifiers to yield a more accurate labeling are proposed and evaluated. These methods lead to higher classification accuracy and provide a mechanism for recognizing deviant signals and false alarms. Performance results are given for signals in the DARPA standard data set I     

Rudder roll stabilization for fishing vessel using neural network approach

This paper presents a neural network (NN) controller for a fishing vessel rudder roll system. The aim of this study is to build a NN controller which uses rudder to regulate both the yaw and roll motion. The neural controller design is accomplished with using the classical back-propagation algorithm (CBA). Effectiveness of the proposed NN control scheme is compared with linear quadratic regulator (LQR) results by simulations carried out a fishing vessel rudder roll stabilizer system.     

An linear matrix inequality approach to global synchronisation of non-parameter perturbations of multi-delay Hopfield neural network

In this study, a successful linear matrix inequality approach is used to analyse a non-parameter perturbation of multi-delay Hopfield neural network by constructing an appropriate Lyapunov-Krasovskii functional. This paper presents the comprehensive discussion of the approach and also extensive applications.     

Tectonic evolution of Gorda Ridge inferred from sidescan sonar images

Gorda Ridge is the southern segment of the Juan de Fuca Ridge complex, in the north-east Pacific. Along-strike spreading-rate variation on Gorda Ridge and deformation of Gorda Plate are evidence for compression between the Pacific and Gorda Plates. GLORIA sidescan sonographs allow the spreading fabric associated with Gorda Ridge to be mapped in detail. Between 5 and 2 Ma, a pair of propagating rifts re-orientated the northern segment of Gorda Ridge by about 10° clockwise, accommodating a clockwise shift in Pacific-Juan de Fuca plate motion that occurred around 5 Ma. Deformation of Gorda Plate, associated with southward decreasing spreading rates along southern Gorda Ridge, is accommodated by a combination of clockwise rotation of Gorda Plate crust, coupled with left-lateral motion on the original normal faults of the ocean crust. Segments of Gorda Plate which have rotated by different amounts are separated by narrow deformation zones across which sharp changes in ocean fabric trend are seen. Although minor lateral movement may occur on these NW to WNW structures, no major right-lateral movement, as predicted by previous models, is observed.     

Interaction of internal waves with a horizontally inhomogeneous density field area overlying a ridge

In the context of the general linear theory, we consider the propagation of an internal tide across a frontal zone overyling an oceanic ridge. For a uniformly stratified ocean, the solution was derived using Riemann's technique. The dependences of the generated internal wave amplitudes on the stratification parameters and bottom topography were determined. We have found that wave disturbances of high intensity inside and in the neighbourhood of the ridge may be concentrated in raytype areas. An increase of the horizontal density gradient in the frontal zone results in a perceptible deformation of these areas.Translated by Vladimir A. Puchkin.     

基于遗传小波神经网络的海底声学底质识别分类

分割海底声纳探测图像,提取单元特征向量进行主成份分析,选取均值、标准差、对比度、相关系数、能量及同质性作为训练特征向量,构建小波神经网络。利用遗传算法优化小波神经网络的初始权值及小波参数,对砂、礁石、泥3种底质类型分别进行训练,并得到3种底质的测试精度都在90%以上,优于单独利用小波神经网络进行训练时的测试精度,克服了小波神经网络训练时易陷入局部极小的固有缺陷,表明基于遗传算法的小波神经网络可有效用于海底底质声纳图像的识别和分类。     

Sea-clutter modeling using a radial-basis-function neural network

Recently, neural networks have been proposed for radar clutter modeling because of the inherent nonlinearity of clutter signals. This paper performs an analysis of the practicality of using a radial basis function (RBF) neural network to model sea clutter and to detect small target embedded in sea clutter. An experiment using an instrumental quality radar was carried out on the eastcoast of Canada to create a rich sea clutter and small surface target database. This database contains both staring and scanning data under various environmental conditions. Using data-sets with different characteristics, we investigate the effects of quantization error, measurement noise, generalization of the neural net over ranges and sampling rate on the RBF clutter model. Despite these physical limitations, the RBF model was shown to approach an optimal predictive performance. The RBF predictor was also applied to detect various small targets in this database based on the constant false alarm rate (CFAR) principle. This RBF-CFAR detector was demonstrated to be able to detect small floating targets even in rough sea conditions     

A neural network for the prediction of wave reflection from coastal and harbor structures

This paper presents the development of an Artificial Neural Network for the prediction of the wave reflection coefficient from a wide range of coastal and harbor structures. The Artificial Neural Network is trained and validated against an extensive database of about 6000 data, including smooth, rock and armor unit slopes, berm breakwaters, vertical walls, low crested structures, oblique wave attacks. The structure and data included in this database, as well as the approach used in this paper, follow the work done on wave overtopping within the CLASH project.In this new Artificial Neural Network 13 input elements are used to represent the physics of the reflection process taking into account the structure geometry (height, submergence, straight or non-straight slope, with or without berm or toe), the structure type (smooth or covered by an armor layer, with permeable or impermeable core) and the wave attack (water depth, wave height, wave length, wave obliquity, directional spreading).The selection of the input elements and of the algorithms used in the network is described based on an in-depth sensitivity analysis of the network performance.The accuracy of the network is quite satisfactory, being the average root mean squared error lower than 0.04. This value is consistent between the Artificial Neural Network calibrated on the original dataset and the one calibrated on boot-strapped datasets in which data reliability and structure complexity are considered.The performance of the network is compared for limited datasets with selected available literature formulae proving that this approach is able to estimate the experimental reflection coefficients with greater accuracy than the empirical formulae calibrated on these same datasets.     

A global seamount classification to aid the scientific design of marine protected area networks

Seamounts are prominent features of the world's seafloor, and are the target of deep-sea commercial fisheries, and of interest for minerals exploitation. They can host vulnerable benthic communities, which can be rapidly and severely impacted by human activities. There have been recent calls to establish networks of marine protected areas on the High Seas, including seamounts. However, there is little biological information on the benthic communities on seamounts, and this has limited the ability of scientists to inform managers about seamounts that should be protected as part of a network. In this paper we present a seamount classification based on "biologically meaningful" physical variables for which global-scale data are available. The approach involves the use of a general biogeographic classification for the bathyal depth zone (near-surface to 3500 m), and then uses four key environmental variables (overlying export production, summit depth, oxygen levels, and seamount proximity) to group seamounts with similar characteristics. This procedure is done in a simple hierarchical manner, which results in 194 seamount classes throughout the worlds oceans. The method was compared against a multivariate approach, and ground-truthed against octocoral data for the North Atlantic. We believe it gives biologically realistic groupings, in a transparent process that can be used to either directly select, or aid selection of, seamounts to be protected.     

人工神经网络在潮汐数值预报中的应用

潮汐数值预报经过了几十年的发展,但是其预报精度并不能让人十分满意,本文试图将传统的潮汐数值预报模式与近年来发展迅速的人工神经网络相结合并改进潮汐数值预报的精度。文章建立了一个神经网络系统,采用潮汐数值模式的输出结果作为网络输入,潮位观测资料作为输出,用建立的神经网络进行训练,结果表明人工神经网络可以明显地改进潮汐数值预报的精度。