Search

This paper compares several new implementations of the YOLO (You Only Look Once) object detection algorithms in harsh underwater environments. Using a dataset collected by a remotely operated vehicle (ROV), we evaluated the performance of YOLOv5, YOLOv6, YOLOv7, and YOLOv8 in detecting objects in challenging underwater conditions. We aimed to determine whether newer YOLO versions are superior to older ones and how much, in terms of object detection performance, for our underwater pipeline dataset. According to our findings, YOLOv5 achieved the highest mean Average Precision (mAP) score, followed by YOLOv7 and YOLOv6. When examining the precision-recall curves, YOLOv5 and YOLOv7 displayed the highest precision and recall values, respectively. Our comparison of the obtained results to those of our previous work using YOLOv4 demonstrates that each version of YOLO detectors provides significant improvement.

Estimating discharge in tidal rivers presents a complex challenge, owing to their dynamic and ever-changing nature. These areas are influenced by multiple factors, including tidal range, wind, atmospheric pressure, and freshwater inflows, causing frequent fluctuations in water levels and discharges. As complex flow patterns and strong turbulence are common for tidal rivers, traditional methods of measuring discharge (such as current meters and stream gauges) are considered inefficient and result in unreliable data. Therefore, more adequate methods are required for estimating discharge. The machine learning approach is one of them. The current study employed a Long Short-Term Memory (LSTM) model that was optimized using grid search cross- validation and applied to simulated data of four water stages and a single discharge station in the Neretva River estuary located in Croatia. The study’s results demonstrated the model’s ability to estimate discharge in low and high-flow conditions accurately. A comparison between LSTM and a simple Multiple Linear Regression (MLR) model clearly shows the improvement in LSTM results for every range of discharge values. This study advances the ability to predict and obtain partial control over discharge in a tidal river by offering crucial insights for more effective environmental and flood management strategies.

The predictive performance of a neural network depends on its weights and architecture. Optimizers based on gradient descent are most commonly used to optimize the weights, and grid search is utilized to find the most suitable architecture from the list of predefined architectures. On the other hand, neuroevolution offers a solution for the simultaneous growth of neural network architecture and the evolution of its weights. Thus, it is not limited by the user- defined list of possible architectures and can find configurations optimal for a specific task. Both approaches can be effectively parallelized and take advantage of modern multi-process systems. In this research, we compare neuroevolution and backpropagation in terms of the time consumed by the algorithm, the predictive performance of the neural network, and the complexity of the neural network. The total time for each algorithm is measured along with the times for each section of the algorithm and the time spent on synchronization due to the multi- process setting. The neural networks are compared by their predictive performance in terms of Matthews correlation coefficient score and their complexity as the number of nodes and connections. The case study is based on two synthetic and two real-world datasets for classification tasks.

While navigating in rough weather, the ship encounters environmental forces and moments such as wind, waves, and sea currents. To maintain eco-efficient ("greener") and safer navigation, appropriate ships' operability and reliability are required. Improvement of ships' performance, and thus its efficiency, can be achieved with an on-board Decision Support System (DSS). The support of such systems is mainly reflected in the lessening of human error, provision of essential data and guidance for operational decisions, as well as the possibility of ship's dynamic positioning with sea waves in areas where operability is limited and prone to error. DSS consists of different integrated modules. A significant contribution is associated with advanced sea state estimation with appropriate signal processing algorithms and methodologies for obtaining data about Directional Wave Spectra (DWS). In recent years, DWS estimation is performed only according to available measured ship motion responses as a more efficient alternative to moored buoys, satellite, and radar measurements. The observed responses are usually the roll, pitch, and heave motions caused by sea waves. However, this so-called buoy analogy approach requires prior knowledge of the ship's performance in particular wave conditions. This information is given in the form of complex transfer functions called Response amplitude operators (RAO), and its accuracy may be insufficient due to incomplete knowledge of input conditions. Accordingly, signal processing and spectral analysis become even more prominent research subjects. Measured ship motion responses are non-stationary signals with time-varying frequency content. Also, these signals are often noisy due to the influence of external forces. Therefore, response signals need to be analysed and filtered to provide accurate information for further processing. For non-stationary signal processing, it is convenient to use Time-Frequency (TF) distribution methods with observation simultaneously in the time and frequency domains. Over the years, different TF methods have been proposed with the common intention of achieving a trade-off between good TF resolution and reduction of undesirable cross-terms in the case of quadratic distributions. The choice of an appropriate method and computer calculation in practical applications depends on the available measured data. Therefore, many adaptive, data-driven TF methods have been developed and proposed, some of them with additional post-processing algorithms. Although such methods provide improved results, estimation bias and variance control issues are often not considered. Besides, the components' resolution may be disturbed with an excessive smoothing process. More desirable results can be obtained by using different Multi-Window TF approaches where multiple window functions are used. These windows are orthogonal (for variance minimization) and optimally concentrated (for low bias estimation). Computational analysis is performed with a given number of window functions by finding optimal weighting coefficients and solving the optimization problem with additional parameters that minimize the variable projection problem. Following this approach, the authors achieved more reliable ship motion analysis results with an indication of possible optimization.