Search

• A novel hybrid approach which integrates a Faster R-CNN for crack patch detection, a DCNN for crack orientation recognition, and a Bayesian algorithm for integration. It provides a novel framework to combine deep learning models and Bayesian analysis to address challenging vision problems where the deep learning approaches with simple end-to-end learning strategy might not be effective. • A distinctive approach to apply Faster R-CNN for the challenging task of crack detection by training it to detect crack patches of suitable SNR, and a semi-automatic method to annotate crack patches of suitable scales to train a Faster R-CNN. • A new Bayesian integration algorithm based on local spatial proximity, orientation consistency and alignment consistency to connect associated neighboring crack patches and suppress false detections, as well as an efficient algorithm to learn the optimal parameters. Vision-based crack detection is of crucial importance in various industries, and it is very challenging due to weak signals in noisy backgrounds. In this paper, we propose a novel hybrid approach for crack detection in raw images, which combines deep learning models and Bayesian probabilistic analysis for robust crack detection. First, we re-train a state-of-the-art object detector (e.g. a Faster R-CNN) to detect crack patches of suitable SNR (signal-noise-ratio). We design a semi-automatic method to generate ground truths of crack patches along crack lines for training. To further improve the accuracy of crack detections over the whole image, we propose a Bayesian integration algorithm to suppress false detections. Specifically, we use a deep CNN to recognize the orientation of the crack segment in each detected patch. Then, a Bayesian probability is computed on the accumulated evidence from detected adjacent patches within a neighborhood based on spatial proximity, orientation consistency and alignment consistency. The patch which lacks local supports is suppressed as false detection. An algorithm to learn the parameters of Bayesian integration is also derived. Extensive experiments and evaluations are performed on a new comprehensive dataset of crack images. The results show that our approach outperforms the state-of-the-art baseline approach on deep CNN classifier. Ablation experiments are also conducted to show the effectiveness of proposed techniques. [ABSTRACT FROM AUTHOR]

• A binary coyote optimization algorithm for feature selection problem is proposed. • The binarization method is based on hyperbolic transfer function. • We test the proposed wrapper with the Naïve Bayes classifier using seven data sets. • The wrapper presents high training accuracy. • The wrapper shows relatively low standard deviation and subsets with few features. The Coyote Optimization Algorithm (COA) is a bio-inspired optimization algorithm based on the intelligent behavior of coyotes. COA was proposed recently and it considers the social organization of the coyotes and its adaptation to the environment in order to solve continuous optimization problems. In addition, it is a population-based algorithm and it can be classified as both, swarm intelligence and evolutionary heuristics, because contributes with a different algorithmic structure. This paper proposes a binary version of the COA, named Binary COA (BCOA) applying to select the optimal feature subset for classification, based on the hyperbolic transfer function in a wrapper model. By this way, the features are selected based on the performance evaluation of a classification algorithm. We tested the effectiveness of the BCOA wrapper with the Naïve Bayes classifier and were used seven public domain benchmark datasets to compare the proposed approach in terms of classification accuracy, number of selected features and computational cost with other state-of-art algorithms of the literature. The results shown that BCOA was able to find subsets with few features while it still performs well in terms of classification accuracy. [ABSTRACT FROM AUTHOR]

• A novel dynamic density peaks clustering method called DPC-DBFN is proposed. • A fuzzy kernel is proposed to compute the local densities of the data points. • A graph-based label propagation strategy is used to identify backbones, border areas and noisy points. • DPC-DBFN can effectively assign true labels to border points located in overlapped regions. • The results on real-world, images and synthetic data show the effectiveness of the proposed method. Density peaks clustering (DPC) is as an efficient clustering algorithm due for using a non-iterative process. However, DPC and most of its improvements suffer from the following shortcomings: (1) highly sensitive to its cutoff distance parameter, (2) ignoring the local structure of data in computing local densities, (3) using a crisp kernel to calculate local densities, and (4) suffering from the cause of chain reaction. To address these issues, in this paper a new method called DPC-DBFN is proposed. The proposed method uses a fuzzy kernel for improving separability of clusters and reducing the impact of outliers. DPC-DBFN uses a density-based kNN graph for labeling backbones. This strategy prevents the chain reaction and effectively assigns true labels to those instances located on the border regions to effectively cluster data with various shapes and densities. The DPC-DBFN is evaluated on some real-world and synthetic datasets. The experimental results show the effectiveness and robustness of the proposed algorithm. [ABSTRACT FROM AUTHOR]

• We propose a TWSVM-based method that includes a kernel enhancement within the TWSVM formulation for imbalanced data classification. • Our proposal, termed enhanced automatic TWSVM' (EATWSVM), allows representing the input samples in a high-dimensional space of possible infinity dimension during the classifier training. • We rule a Gaussian-based similarity through a Mahalanobis distance to learn the EATWSVM's kernel function using a centered kernel alignment (CKA) -based method. • We suggest a suitable range to fix the regularization parameters concerning both the dataset imbalanced ratio and overlap. • EATWSVM obtains a remarkable trade-off between classification performance and training time, without the need of a sophisticated prior user knowledge concerning the algorithm tuning. Most of the classification approaches assume that the sample distribution among classes is balanced. Still, such an assumption leads to biased performance over the majority class. This paper proposes an enhanced automatic twin support vector machine – (EATWSVM) to deal with imbalanced data, which incorporates a kernel representation within a TWSVM-based optimization. To learn the kernel function, we impose a Gaussian similarity, ruled by a Mahalanobis distance, and couple a centered kernel alignment-based approach to improving the data separability. Besides, we suggest a suitable range to fix the regularization parameters concerning both the dataset' imbalance ratio and overlap. Lastly, we adopt One-vs-One and One-vs-Rest frameworks to extend our EATWSVM formulation for multi-class tasks. Obtained results on synthetic and real-world datasets show that our approach outperforms state-of-the-art methods concerning classification performance and training time. [ABSTRACT FROM AUTHOR]

• Hybridizing Flower Pollination with hill climbing (FPAc) for EEG-based person identification. • FPAc is used to select the optimal EEG channels which can provides high accuracy. • FPAc tested using standard EEG dataset, namely, EEG motor movement/imagery dataset. • The results of FPAc are compared with standard FPA and with other state-of-arts. • Effect of FPAc on the performance of EEG channels selection is studied and shows significant improvements. Recently, electroencephalogram (EEG) signal presents a great potential for a new biometric system to deal with a cognitive task. Several studies defined the EEG with uniqueness features, universality, and natural robustness that can be used as a new track to prevent spoofing attacks. The EEG signals are the graphical recording of the brain electrical activities which can be measured by placing electrodes (channels) in various positions of the scalp. With a large number of channels, some channels have very important information for biometric system while others not. The channel selection problem has been recently formulated as an optimisation problem and solved by optimisation techniques. This paper proposes hybrid optimisation techniques based on binary flower pollination algorithm (FPA) and β -hill climbing (called FPA β -hc) for selecting the most relative EEG channels (i.e., features) that come up with efficient accuracy rate of personal identification. Each EEG signals with three different groups of EEG channels have been utilized (i.e., time domain, frequency domain, and time-frequency domain). The FPA β -hc is measured using a standard EEG signal dataset, namely, EEG motor movement/imagery dataset with a real world data taken from 109 persons each with 14 different cognitive tasks using 64 channels. To evaluate the performance of the FPA β -hc, five measurement criteria are considered:accuracy (Acc), (ii) sensitivity (Sen), (iii) F-score (F_s), (v) specificity (Spe), and (iv) number of channels selected (No. Ch). The proposed method is able to identify the personals with high Acc, Sen., F_s, Spe, and less number of channels selected. Interestingly, the experimental results suggest that FPA β -hc is able to reduce the number of channels with accuracy rate up to 96% using time-frequency domain features. For comparative evaluation, the proposed method is able to achieve results better than those produced by binary-FPA-OPF method using the same EEG motor movement/imagery datasets. In a nutshell, the proposed method can be very beneficial for effective use of EEG signals in biometric applications. [ABSTRACT FROM AUTHOR]

• We summarize the binary neural network methods and categorize them into the naive binarization and the optimized binarization. • The binary neural networks are mainly optimized using techniques including minimizing quantization error, improving the loss function, and reducing the gradient error. • We also discuss the hardware-friendly methods and the useful tricks of training binary neural networks. • We present the common datasets and network structures of evaluation, and compare the performance on different tasks. • We conclude and point out the future research trends. The binary neural network, largely saving the storage and computation, serves as a promising technique for deploying deep models on resource-limited devices. However, the binarization inevitably causes severe information loss, and even worse, its discontinuity brings difficulty to the optimization of the deep network. To address these issues, a variety of algorithms have been proposed, and achieved satisfying progress in recent years. In this paper, we present a comprehensive survey of these algorithms, mainly categorized into the native solutions directly conducting binarization, and the optimized ones using techniques like minimizing the quantization error, improving the network loss function, and reducing the gradient error. We also investigate other practical aspects of binary neural networks such as the hardware-friendly design and the training tricks. Then, we give the evaluation and discussions on different tasks, including image classification, object detection and semantic segmentation. Finally, the challenges that may be faced in future research are prospected. [ABSTRACT FROM AUTHOR]

• We generalise zipping algorithms for segmentation hierarchy editing to simplify the popular TugGraph visualisation algorithm. • Our SimpleTug approach is modular, straightforward, and saves memory and time compared to existing TugGraph implementations. • We prove its equivalence to other methods, show its effectiveness, and discuss how TugGraph and zipping are connected. • These connections help pin down what TugGraph is really doing, making it easier to reason about in future. • We further show how SimpleTug and an existing TugGraph method can be parallelised. Graphs are an invaluable modelling tool in many domains, but visualising large graphs in their entirety can be difficult. Hierarchical graph visualisation – recursively clustering a graph's nodes to view it at a higher level of abstraction – has thus become popular. However, this can hide important information that a user needs to understand a graph's topology, e.g. nodes' neighbourhoods. TugGraph addressed this by 'separating out' a given node's neighbours from their hierarchy ancestors to visualise them independently. Its original implementation was straightforward, but copied parts of the hierarchy, making it slow and memory-hungry. An optimised later version, which we refer to as FastTug , avoided this, but at a cost in clarity. Optimising TugGraph without sacrificing clarity is difficult because of the need to keep every hierarchy node connected, a common challenge for graph hierarchy editing algorithms. Recently, this problem has been addressed by 'zipping' algorithms, multi-level split/merge algorithms that preserve hierarchy node connectedness and can be built upon for higher-level editing. In this paper, we generalise the original unzipping algorithms to implement SimpleTug , a simple, modular version of TugGraph that is easy to understand and implement, and even faster and more memory-efficient than FastTug. We formally prove its equivalence to FastTug , and show how both can be parallelised. Using our millipede hierarchical image segmentation system, we show experimentally that both the serial and parallel versions of SimpleTug are around 25% faster than their FastTug counterparts, whilst using considerably less memory. Finally, we discuss the interesting theoretical connections between TugGraph and zipping, and suggest ideas for further work. [ABSTRACT FROM AUTHOR]

• Nearest neighbor graph can indicate the samples that lying within the local dense regions of dataset without any input parameter. • A clustering algorithm named ADBSCAN is developed based on the nearest neighbor graph properties. • Experiments on different types of datasets demonstrate the superior performance and the robust to parameters of ADBSCAN. Density-based clustering has several desirable properties, such as the abilities to handle and identify noise samples, discover clusters of arbitrary shapes, and automatically discover of the number of clusters. Identifying the core samples within the dense regions of a dataset is a significant step of the density-based clustering algorithm. Unlike many other algorithms that estimate the density of each samples using different kinds of density estimators and then choose core samples based on a threshold, in this paper, we present a novel approach for identifying local high-density samples utilizing the inherent properties of the nearest neighbor graph (NNG). After using the density estimator to filter noise samples, the proposed algorithm ADBSCAN in which "A" stands for "Adaptive" performs a DBSCAN-like clustering process. The experimental results on artificial and real-world datasets have demonstrated the significant performance improvement over existing density-based clustering algorithms. [ABSTRACT FROM AUTHOR]

• An ensemble selection method that takes into account each base classifier's confidence during classification and its overall credibility on the task is proposed. • The overall credibility of a base classifier is obtained by minimizing the empirical 0–1 loss on the entire training set. • The classifier's confidence in prediction for a test sample is measured by the entropy of its soft classification outputs for that sample. • Extensive comparative experiments with the state-of-the-art algorithms on ensemble selection validated the superior performance of our algorithm. Ensemble selection is one of the most studied topics in ensemble learning because a selected subset of base classifiers may perform better than the whole ensemble system. In recent years, a great many ensemble selection methods have been introduced. However, many of these lack flexibility: either a fixed subset of classifiers is pre-selected for all test samples (static approach), or the selection of classifiers depends upon the performance of techniques that define the region of competence (dynamic approach). In this paper, we propose an ensemble selection method that takes into account each base classifier's confidence during classification and the overall credibility of the base classifier in the ensemble. In other words, a base classifier is selected to predict for a test sample if the confidence in its prediction is higher than its credibility threshold. The credibility thresholds of the base classifiers are found by minimizing the empirical 0–1 loss on the entire training observations. In this way, our approach integrates both the static and dynamic aspects of ensemble selection. Experiments on 62 datasets demonstrate that the proposed method achieves much better performance in comparison to some ensemble methods. [ABSTRACT FROM AUTHOR]