Search

This paper proposes a three-stage online deep learning model for time series based on the ensemble deep random vector functional link (edRVFL). The edRVFL stacks multiple randomized layers to enhance the single-layer RVFL's representation ability. Each hidden layer's representation is utilized for training an output layer, and the ensemble of all output layers forms the edRVFL's output. However, the original edRVFL is not designed for online learning, and the randomized nature of the features is harmful to extracting meaningful temporal features. In order to address the limitations and extend the edRVFL to an online learning mode, this paper proposes a dynamic edRVFL consisting of three online components, the online decomposition, the online training, and the online dynamic ensemble. First, an online decomposition is utilized as a feature engineering block for the edRVFL. Then, an online learning algorithm is designed to learn the edRVFL. Finally, an online dynamic ensemble method, which can measure the change in the distribution, is proposed for aggregating all layers' outputs. This paper evaluates and compares the proposed model with state-of-the-art methods on sixteen time series. [ABSTRACT FROM AUTHOR]

• We propose an automated layer-wise pipeline for randomized feed-forward neural networks. • We propose a Sample-level Attention Strategy to solve the capabilities problem. • Several ablation experiments are conducted to verify the effect of hyper-parameters. The randomized feed-forward neural network is a single hidden layer feed-forward neural network that enables efficient learning by optimizing only the output weights. The ensemble deep learning framework significantly improves the performance of randomized neural networks. However, the framework's capabilities are limited by traditional hyper-parameter selection approaches. Meanwhile, different random network architectures, such as the existence or lack of a direct link and the mapping of direct links, can also strongly affect the results. We present an automated learning pipeline for the ensemble deep randomized feed-forward neural network in this paper, which integrates hyper-parameter selection and randomized network architectural search via Bayesian optimization to ensure robust performance. Experiments on 46 UCI tabular datasets show that our strategy produces state-of-the-art performance on various tabular datasets among a range of randomized networks and feed-forward neural networks. We also conduct ablation studies to investigate the impact of various hyper-parameters and network architectures. [ABSTRACT FROM AUTHOR]

The volatile characteristics of the tanker market pose challenges to forecasting. In addition, the volatile characteristics of newbuilding and secondhand ship prices, time charter rates, and scrap values make developing a unified framework of forecasting difficult. Most researchers have developed forecasting models and evaluated their performance based on a specific market. Such narrow development imposes difficulty for practitioners to choose a suitable model. Due to the boom of machine learning, many researchers are trying to boost the forecasting accuracy of shipping markets using machine learning. However, there are many hyper-parameters of the complex machine learning models and a slight variation of the model may cause significant performance degradation. This paper utilizes a forecast combination mechanism to forecast many time series collected from the shipping market, including newbuilding and secondhand ship prices, scrap values, and time charter rates. The models inside the combination pool are just linear functions. Finally, we compare their performance with conventional machine learning models and naïve forecasts using three error metrics and statistical tests. The statistical tests show that the combination of linear models is superior. The findings of this study also indicate that complex models do not boost forecasting accuracy necessarily. [ABSTRACT FROM AUTHOR]

The security and reliability of communication systems are crucial for Intelligent Transport Systems (ITS), ensuring user data privacy and the transportation network's stable operation. To address challenges where certain Vehicle User Equipment (VUE) cannot establish stable message exchange with the Base Station (BS), we employ Internet of Things (IoT) nodes to assist in communication setup. Specifically, we propose a weighted fractional Fourier transform (WFRFT)-based two-way relay physical layer security (PLS) communication scheme in the signal domain for IoT-aided ITS. The BS and VUE encode confidential messages into WFRFT signals with private transform orders and perfectly decode the received signal in the specific signal domain. However, in the decoding results at the eavesdropper, not only is the energy of information-bearing signals reduced, but the lost energy also transforms into inter-component interference, further degrading the quality of its received signal. The security gains stem from the mismatched signal domain of the eavesdropper and do not require additional transmit power. Moreover, the security performance is detailed, and the secrecy rate in a time-varying channel is derived. Finally, numerical results validate the theoretical analysis and demonstrate the performance of the proposed scheme. • Propose a two-way relay PLS communication system for ITS. • Propose a WFRFT-based PLS scheme in the signal domain. • The security performance is analyzed theoretically and substantiated. [ABSTRACT FROM AUTHOR]

Efficient forecasting of ship order books holds immense significance in the maritime industry, enabling companies to optimize their operations, allocate resources effectively, and make informed decisions. However, volatile characteristics within historical order books pose challenges in achieving reliable, intelligent, and precise forecasts. This paper presents a novel ensemble deep random vector functional link (edRVFL) algorithm to anticipate future ship order book dynamics. The edRVFL leverages deep feature extraction and ensemble learning to enhance forecasting performance. To further elevate its capabilities, we introduce a discontinuous and parsimonious embedding strategy, which deviates from the conventional dense collection of continuous time steps used in vanilla edRVFL. This parsimonious embedding approach limits the model's complexity and boosts its generalization ability. We extensively evaluate the proposed method using ship order book data, and comparative studies demonstrate its superiority over alternative approaches. Our proposed edRVFL offers a promising solution for accurate and efficient ship order book forecasting, making it a valuable asset in the maritime industry's decision-making processes. The source codes utilized in this research are openly available on GitHub at the following link: https://github.com/crkkkaa/Ship-order-book-forecasting-by-an-ensemble-deep-parsimonious-random-vector-functional-link-network-. • A novel ensemble deep RVFL to anticipate ship order book dynamics is proposed. • We propose a parsimonious embedding algorithm to optimize temporal patterns. • The direct links ablation study is conducted for comparison. [ABSTRACT FROM AUTHOR]

Efficient forecasting of ship order books holds immense significance in the maritime industry, enabling companies to optimize their operations, allocate resources effectively, and make informed decisions. However, volatile characteristics within historical order books pose challenges in achieving reliable, intelligent, and precise forecasts. This paper presents a novel ensemble deep random vector functional link (edRVFL) algorithm to anticipate future ship order book dynamics. The edRVFL leverages deep feature extraction and ensemble learning to enhance forecasting performance. To further elevate its capabilities, we introduce a discontinuous and parsimonious embedding strategy, which deviates from the conventional dense collection of continuous time steps used in vanilla edRVFL. This parsimonious embedding approach limits the model's complexity and boosts its generalization ability. We extensively evaluate the proposed method using ship order book data, and comparative studies demonstrate its superiority over alternative approaches. Our proposed edRVFL offers a promising solution for accurate and efficient ship order book forecasting, making it a valuable asset in the maritime industry's decision-making processes. The source codes utilized in this research are openly available on GitHub at the following link: https://github.com/crkkkaa/Ship-order-book-forecasting-by-an-ensemble-deep-parsimonious-random-vector-functional-link-network-. • A novel ensemble deep RVFL to anticipate ship order book dynamics is proposed. • We propose a parsimonious embedding algorithm to optimize temporal patterns. • The direct links ablation study is conducted for comparison. [ABSTRACT FROM AUTHOR]

• A novel PIMSE approach is proposed to forecast second-hand oil tanker price. • The proposed algorithm ensures a parsimonious lag structure while maintaining high interpretability of the selected lags and explanatory variables. • The proposed model outperforms the baselines in terms of three error metrics on three time series from the shipping market. Estimating second-hand ship prices in the highly uncertain and cyclical ship trading market is a challenge due to its volatile nature. In this study, we propose a novel and highly interpretable model termed as the parsimonious intelligent model search engine (PIMSE), to investigate the relationship between key supply variables and the prices of second-hand oil tankers. Through empirical evaluation using a time series dataset spanning from 2002 to 2020, encompassing three types of oil tankers (VLCC, Suezmax, and Aframax), we assess the effectiveness and performance of PIMSE. The results demonstrate the superior performance of PIMSE compared to other estimation models in terms of its accuracy and stability. It can effectively address abrupt structural change and capture trend and seasonal variations in the time series data. Moreover, the high interpretability of PIMSE provides valuable insights into the factors that influence second-hand ship prices, empowering stakeholders in the shipping industry to make well-informed investment decisions. By leveraging PIMSE, decision-makers can gain a deeper understanding of market dynamics and navigate the complexities of the ship trading industry. Notably, PIMSE's ability to handle the cyclical nature of the ship trading market and incorporate trend and seasonal variations enhances the robustness and accuracy of second-hand ship price estimation. [ABSTRACT FROM AUTHOR]

Randomized neural networks (RNNs) have shown outstanding performance in many different fields. The superiority of having fewer training parameters and closed-form solutions makes them popular in small datasets analysis. However, automatically decoding raw electroencephalogram (EEG) data using RNNs is still challenging in EEG-based passive brain–computer interface (pBCI) classification tasks. Models with the high-dimension input of EEG may suffer from overfitting and the intrinsic characteristics of non-stationary, high-level noises and subject variability could limit the generation of distinctive features in the hidden layers. To address these problems in EEG-based pBCI tasks, this work proposes a spectral-ensemble deep random vector functional link (SedRVFL) network that focuses on feature learning in the frequency domain. Specifically, an unsupervised feature-refining (FR) block is proposed to improve the low feature learning capability in RNNs. Moreover, a dynamic direct link (DDL) is performed to further complement the frequency information. The proposed model has been evaluated on a self-collected dataset as well as a public driving dataset. The cross-subject classification results obtained demonstrated its effectiveness. This work offers a new solution for EEG decoding, i.e. , using optimized RNNs for decoding complex raw EEG data and boosting the classification performance of EEG-based pBCI tasks. • A SedRVFL network is proposed for the decoding of high-dimension raw EEG data. • A FR block is presented to improve the feature learning capability of the edRVFL. • Non-linear features are employed in the direct link. • A dynamic direct link algorithm is proposed to reduce the redundant information. [ABSTRACT FROM AUTHOR]

This work investigated the use of an ensemble deep random vector functional link (edRVFL) network for electroencephalogram (EEG)-based driver fatigue recognition. Against the low feature learning capability of the edRVFL network from raw EEG signals, two strategies were exploited in this work. Specifically, the first one was to exploit the advantages of the feature extractor module in CNNs, i.e. , use CNN features as the input of the edRVFL network. The second one was to improve the feature learning capability of the edRVFL network. An enhanced edRFVL network named FGloWD-edRVFL was proposed, in which four enhancements were implemented, including random forest-based F eature selection, Glo bal output layer, W eighting and entropy-based D ynamic ensemble. The proposed FGloWD-edRVFL network was evaluated on the challenging cross-subject driver fatigue recognition tasks. The results indicated that the proposed model could boost the recognition performance, significantly outperforming all strong baselines. The step-wise analysis further demonstrated the effectiveness of the proposed enhancements in the edRVFL network. • An enhanced edRVFL input with CNN features is proposed. • RF-based feature selection is used to remove inferior enhancement features. • Weighting and a global output layer are implemented to diversify ensemble learning. • A dynamic ensemble is used to reduce the impact of low-competence classifiers. [ABSTRACT FROM AUTHOR]

The reliable control of wave energy devices highly relies on the forecasts of wave heights. However, the dynamic characteristics and significant fluctuation of waves' historical data pose challenges to precise predictions. Neural networks offer a promising solution to forecast the wave heights by extracting meaningful features from historical observations. This paper proposes a novel hybrid random vector functional link network with the ensemble and deep learning benefits. Hierarchical stacks of hidden layers are constructed to enforce the deep representations of the time series. Individual output layers follow all enhancement layers to adopt ensemble learning. A neuron pruning strategy is proposed to remove the noisy information from the random features and boost the network's performance. Besides, the proposed network is further utilized to forecast the additive and multiplicative residuals from the ARIMA method. Finally, the ensemble of additive-ARIMA-edRVFL, multiplicative-ARIMA-edRVFL, and edRVFL achieves the best average rankings around two for three forecasting horizons. The proposed ensemble achieves an average ranking of 1.33 on four-hours ahead of forecasting in terms of root mean square error and mean absolute scaled error. Extensive experiments are conducted on twelve time series of the significant wave height. The comparative results demonstrate the superiority of the proposed model over other state-of-the-art methods. The source codes are available on https://github.com/P-N-Suganthan/CODES. [ABSTRACT FROM AUTHOR]

Forecasts of the wave heights can assist in the data-driven control of wave energy systems. However, the dynamic properties and extreme fluctuations of the historical observations pose challenges to the construction of forecasting models. This paper proposes a novel dynamic ensemble deep Echo state networks (ESN) to learn the dynamic characteristics of the significant wave height. The dynamic ensemble ESN creates a profound representation of the input and trains an independent readout module for each reservoir. To begin, numerous reservoir layers are built in a hierarchical order, adopting a reservoir pruning approach to filter out the poorer representations. Finally, a dynamic ensemble block is used to integrate the forecasts of all readout layers. The suggested model has been tested on twelve available datasets and statistically outperforms state-of-the-art approaches. • A dynamic ensemble deep echo state network is proposed. • A reservoir pruning strategy is proposed to remove inferior features. • An extensive review of significant wave height forecasting is conducted. • Our model manifests better performance than the SOTA models. [ABSTRACT FROM AUTHOR]

Electric load forecasting is essential for the planning and maintenance of power systems. However, its un-stationary and non-linear properties impose significant difficulties in predicting future demand. This paper proposes a novel ensemble deep Random Vector Functional Link (edRVFL) network for electricity load forecasting. The weights of hidden layers are randomly initialized and fixed during the training process. The hidden layers are stacked to enforce deep representation learning. Then, the model generates the forecasts using the ensemble of the outputs of each layer. Moreover, we also propose to augment the random enhancement features by empirical wavelet transformation (EWT). The raw load data are decomposed by EWT in a walk-forward approach without introducing future data leakage problems in the decomposition process. Finally, all the sub-series generated by the EWT, including raw data, are fed into the edRVFL for forecasting purposes. The proposed model is evaluated on sixteen publicly available time series from the Australian Energy Market Operator of the year 2020. The simulation results demonstrate the proposed model's superior performance over eleven forecasting methods in two error metrics and statistical tests on electricity load forecasting tasks. • An ensemble deep randomized network for forecasting is proposed. • Empirical wavelet transformation is used to augment unsupervised features. • A layer-wise tuning algorithm is proposed to optimize the structures. • Two error metrics and statistical tests are utilized to compare the models. [ABSTRACT FROM AUTHOR]

Emotion recognition based on electroencephalogram (EEG) signals is helpful in various fields, including medical healthcare. One possible medical application is to diagnose emotional disorders in patients. Humans tend to work and communicate efficiently when in a good mood. On the other hand, negative emotions can harm physical and mental health. Traditional EEG-based methods usually extract time-domain and frequency-domain features before classifying them. Convolutional Neural Networks (CNN) enables us to extract features and classify them end-to-end. However, most CNN methods use backpropagation to train their models, which can be computationally expensive, primarily when a complex model is used. Inspired by the successes of Random Vector Functional Link and Convolutional Random Vector Functional Link, we propose using a randomized CNN model for emotion recognition that removes the need for a backpropagation method. Also, we expand our randomized CNN method to a deep and ensemble version to improve emotion recognition performance. We do experiments on the commonly used publicly available Database for Emotion Analysis using the Physiological Signals (DEAP) dataset to evaluate our randomized CNN models. Results on the DEAP dataset show our models outperform all other models, with at least 95% accuracy for all subjects. Our ensemble version outperforms our shallow version, winning the shallow version in most subjects. • A randomized CNN is proposed to classify EEG signals. • Randomized CNN are expanded into deep and ensemble version for improved performance. • Randomized CNN outperforms other EEG-based models. • Deep and ensemble version substantially outperforms shallow Randomized CNN. [ABSTRACT FROM AUTHOR]

Asset prices play a significant role in the financial survival and profitability of ship-owning firms. In a highly volatile shipping market, prices of newbuilding ships must be predicted to detect security shortfalls as well as opportunities for temporal arbitration (gaining on high–low pricing). Accordingly, this paper proposes an improved version of the intelligent model search engine (IMSE) by asynchronous time lag selection. The parsimonious IMSE algorithm comprises the essential components such as input and training data size selection by a grid search procedure. In the initial IMSE algorithm, time-lag (memory size) selection is designed such that a serial cluster of memory groups is assigned synchronously for all inputs. By relaxing of lag structures selection, the proposed algorithm estimates unique lead–lag relations for the input of the intended problem set. An extensive benchmark study with several baseline models and the persistence forecast (Naïve I) is performed to observe the out-of-sample accuracy of the proposed approach. The empirical results indicate that second-hand ship prices, scrap values, and orderbook (no. of orders) have predictive features and are selected by the search engine for two ship sizes. Different lag structures are estimated for each input with asynchronous time-lag selection improvement. • A novel intelligent model search engine algorithm is proposed for newbuilding ship prices. • The proposed algorithm ensures the parsimony of the lag structure. • The selected lags and explanatory variables are highly interpretable. [ABSTRACT FROM AUTHOR]

Anticipating newbuilding ship prices is crucial for participants in the dynamic shipping market. Although the researchers from forecasting and shipping have shown that the machine learning models outperform statistical ones, convolution neural networks are not investigated. The convolution neural networks are proposed for image processing, rendering difficulty when handling monthly time series. This paper presents a light neural network with annual dilated convolution filters while extracting the newbuilding market’s short-term and long-term temporal knowledge. The multivariate shipping data are fed into multiple convolutional filters with nonlinear activations. Finally, the convoluted features are fed into a linear layer which maps the features to future values. The annual dilated convolution filter owns a vision across one year and integrates all variables’ temporal information. Besides, the dilation rate renders a parsimonious structure, preventing the model from overfitting. The proposed model is compared with statistical models, Naïve forecasts, and various machine learning models on the newbuilding prices of three tanker markets. The empirical results highlight the superiority of the proposed convolutional neural networks. [ABSTRACT FROM AUTHOR]

The lack of intuitive and active human–robot interaction makes it difficult to use upper-limb-assistive devices. In this paper, we propose a novel learning-based controller that intuitively uses onset motion to predict the desired end-point position for an assistive robot. A multi-modal sensing system comprising inertial measurement units (IMUs), electromyographic (EMG) sensors, and mechanomyography (MMG) sensors was implemented. This system was used to acquire kinematic and physiological signals during reaching and placing tasks performed by five healthy subjects. The onset motion data of each motion trial were extracted to input into traditional regression models and deep learning models for training and testing. The models can predict the position of the hand in planar space, which is the reference position for low-level position controllers. The results show that using IMU sensor with the proposed prediction model is sufficient for motion intention detection, which can provide almost the same prediction performance compared with adding EMG or MMG. Additionally, recurrent neural network (RNN)-based models can predict target positions over a short onset time window for reaching motions and are suitable for predicting targets over a longer horizon for placing tasks. This study's detailed analysis can improve the usability of the assistive/rehabilitation robots. [ABSTRACT FROM AUTHOR]

• Evolutionary algorithms cause massive computations in selecting lagged variables. • Supervised PCA is an approach to solve the lag-length optimization problem. • The redundant left-hand side is transformed into a concise one by supervised PCA. Lag-selection is a high dimensional hyper-parameter in the fuzzy time series (FTS) which requires complex optimization process and computational capacity particularly in high frequency dataset (e.g. daily, hourly). Multivariate high order FTS suffers from establishing long logical relationships, and the difficulty of rule matching is proportional to the time lags and number of variables. In the vast majority of FTS literature, a grid search algorithm or evolutionary algorithms are run to find singular time-lags. In addition, some researchers determine the lag structure arbitrarily. However, grid search in high dimensional problems is not practical especially when recursive predictions are generated and evolutionary algorithms suffer from the randomness which may generate different solutions. This paper proposes an alternative approach to the lag selection problem by utilizing supervised principal component analysis (SPCA), and the lag structure is reduced to low dimensional space. SPCA has been developed to project the high dimensional lagged variables into the first principal component. An empirical study is conducted to validate the proposed approach by using global shipping industry data in the world. [ABSTRACT FROM AUTHOR]

General deep learning methods achieve high-level semantic feature representation by aggregating hierarchical features, which performs well in object detection tasks. However, issues arise with general deep learning methods in UAV-based remote sensing image object detection tasks. Firstly, general feature aggregation methods such as stride convolution may lead to information loss in input samples. Secondly, common FPN methods introduce conflicting information by directly fusing feature maps from different levels. These shortcomings limit the model's detection performance on small and weak targets in remote sensing images. In response to these concerns, we propose an unmanned aerial vehicle (UAV) object detection algorithm, IF-YOLO. Specifically, our algorithm leverages the Information-Preserving Feature Aggregation (IPFA) module to construct semantic feature representations while preserving the intrinsic features of small objects. Furthermore, to filter out irrelevant information introduced by direct fusion, we introduce the Conflict Information Suppression Feature Fusion Module (CSFM) to improve the feature fusion approach. Additionally, the Fine-Grained Aggregation Feature Pyramid Network (FGAFPN) facilitates interaction between feature maps at different levels, reducing the generation of conflicting information during multi-scale feature fusion. The experimental results on the VisDrone2019 dataset demonstrate that in contrast to the standard YOLOv8-s, our enhanced algorithm achieves a mean average precision (mAP) of 47.3%, with precision and recall rates enhanced by 6.3% and 5.6%, respectively. [ABSTRACT FROM AUTHOR]

Fuzzy cognitive maps have achieved significant success in time series modeling and forecasting. However, fuzzy cognitive maps still contain weakness to handle the nonstationarity and outliers. We propose a novel time series forecasting model based on fuzzy cognitive maps and empirical wavelet transformation in this paper. The empirical wavelet transformation is applied to decompose the original time series into different levels which capture information of different frequencies. Then, the high-order fuzzy cognitive map is trained to model the relationships among all the sub-series generated and original time series. To enhance the robustness of high-order fuzzy cognitive maps against outliers, a novel learning method based on support vector regression is designed. Finally, we divide the summation of each concept value of the high-order fuzzy cognitive map by two to obtain the numerical predictions. A comprehensive empirical study on eight public time series validates the superiority of proposed model compared with the popular baseline models from the literature. • A novel EWT-FCM is proposed to time series forecasting. • The EWT is developed to decompose the original time series. • A robust learning algorithm is designed for high-order FCM. • Proposed model is validated through an empirical study of various time series. [ABSTRACT FROM AUTHOR]

• Establish parsimonious fuzzy time series model objectively and automatically. • Propose a novel complexity term for fuzzy time series model. • Integrate data transformation, feature selection and hyper-parameter optimization together. • A two-stage GA optimization is proposed to achieve sparsi cation and accuracy simultaneously. This paper proposes a novel modelling structure to ensure the parsimony of fuzzy time series (FTS) models while retaining certain level of out-of-sample accuracy. A parsimonious FTS model requires multiple optimizations of hyper-parameters such as time lags and partitioning which consists of the number of fuzzy sets, the partitioning type and the membership functions. In the vast literature of fuzzy time series, hyper-parameter optimization is usually ignored. In addition to that, optimization process for the hyper-parameters is also not presented properly. In this study, a parsimonious FTS modelling approach is introduced by using genetic algorithm (GA). Three major innovations are proposed: (1) Hyper-parameters of FTS structure are optimized to eliminate subjective preferences with the help of GA. Some of those parameters are never optimized or simply ignored in the past research. (2) The set of hyper-parameters is optimized subject to highest accuracy in validation set data and model's complexity. (3) For achieving sparsification and accuracy simultaneously at reasonable computation time, a two-stage GA optimization is run to search for higher accuracy and lower complexity consecutively. Empirical studies are conducted on two types of datasets. Prices of liquid bulk cargo carriers (i.e. tanker) and secondhand ship have been predicted using the proposed approach. Potential benchmarks as well as a simple Nave forecast have been compared to the proposed model for validation based on mean absolute scaled error and root mean squared error. [ABSTRACT FROM AUTHOR]

With the mandatory implementation of the automatic identification system and the rapid advancement of relevant satellite communication technologies, a vast amount of vessel trajectory data has been amassed. It has catalyzed advancements in the field of maritime anomaly detection, significantly contributing to the enhancement of intelligent maritime situational awareness. However, detecting abnormal vessel trajectories from massive data is a highly challenging task that requires extensive manual effort. To address the challenge, this paper develops an unsupervised deep learning method for detecting abnormal vessel trajectories. Specifically, we first generate the informative trajectory images by remapping the raw vessel trajectories into two-dimensional matrices while maintaining the spatio-temporal properties. Then, a Wasserstein generative adversarial network with gradient penalty (WGAN-GP) is trained on normal vessel trajectories. Next, an encoder is trained to map these trajectory images to a latent space. Finally, the trained encoder and WGAN-GP are used to design an anomaly score for detecting abnormal vessel trajectories and anomalies in vessel trajectories. Experimental evaluations using multiple evaluation metrics on diverse simulated and real-world maritime environments validate the efficacy of the proposed method in unsupervised maritime anomaly detection. The results indicate that the proposed model can provide accurate anomaly detection on vessel trajectories without manual labeling. [ABSTRACT FROM AUTHOR]

With the advancement of maritime unmanned aerial vehicles (UAVs) and deep learning technologies, the application of UAV-based object detection has become increasingly significant in the fields of maritime industry and ocean engineering. Endowed with intelligent sensing capabilities, the maritime UAVs enable effective and efficient maritime surveillance. To further promote the development of maritime UAV-based object detection, this paper provides a comprehensive review of challenges, relative methods, and UAV aerial datasets. Specifically, in this work, we first briefly summarize four challenges for object detection on maritime UAVs, i.e., object feature diversity, device limitation, maritime environment variability, and dataset scarcity. We then focus on computational methods to improve maritime UAV-based object detection performance in terms of scale-aware, small object detection, view-aware, rotated object detection, lightweight methods, and others. Next, we review the UAV aerial image/video datasets and propose a maritime UAV aerial dataset named MS2ship for ship detection. Furthermore, we conduct a series of experiments to present the performance evaluation and robustness analysis of object detection methods on maritime datasets. Eventually, we give the discussion and outlook on future works for maritime UAV-based object detection. The MS2ship dataset is available at https://github.com/zcj234/MS2ship. [ABSTRACT FROM AUTHOR]

Deep learning and fuzzy models provide powerful and practical techniques for solving large-scale deep-learning tasks. The fusion technique on deep learning and fuzzy system are generally classified into ensemble and integrated modes and materializes in information fusion, model fusion, and feature fusion. In an ensemble-based fusion, the fuzzy model either acts as an activation function or is operated as a separate process aggregating/preprocessing the information. Some early attempts in the field have successfully fused deep neural networks and fuzzy modeling concepts in ensemble mode. However, no effective attempts were made to fuse fuzzy models as an integrated feature-level fusion learning with graph neural networks (GNNs). This is mainly due to two challenges related to this fusion: (1) the number of fuzzy rules grows exponentially with the number of features that causes computational inefficiency, and (2) the solution space created by this fusion of fuzzy rules becomes complex due to multiple regression relations between inputs and outputs. Additionally, a simple linear regression at the output space would not be sufficient to model deep learning tasks. Therefore, this paper addresses these challenges by proposing a feature-level fusion method to fuse deep learning and fuzzy modeling where the latter technique is for integrated feature learning, called fuzzy forest graph neural network (FuzzyGNN), which creates a fuzzy learning forest fusing the linear graph transformers for deep learning tasks. We conducted experiments on fourteen machine learning datasets to test and validate the efficiency of the proposed FuzzyGNN model. Compared to state-of-the-art methods, our algorithm achieves the best results on four out of five machine learning datasets. The source code will be available at https://github.com/lingping-fuzzy/ , https://github.com/vojha-code and https://github.com/P-N-Suganthan. • We present an integrated fusion method for large-scale dataset classification. • We propose an integrated fusion of fuzzy learning and graph neural networks. • We propose sampling methods to address fuzzy rules exponentially explosion problem. • We propose adopting polynomial functions to simulate fitting prediction. • We experimentally justify the performance of the proposed model. [ABSTRACT FROM AUTHOR]