Benchmark Databases Research Articles

An adaptive weighting multimodal fusion classification system for steel plate surface defect

In steel surface inspection, an accurate steel surface defect identification method is needed to evaluate the impact of defects on structural performance and system maintenance. Traditionally, the recognition accuracy of methods based on handcrafted features is limited, but the system performance can be improved by feature fusion extracted by different methods. Therefore, this research uses the pre-trained convolutional neural network (CNN) combined with transfer learning to extract effective abstract features, and carries out adaptive weighting multimodal fusion of three the abstract features and handcrafted feature sets at the decision-making level, that is, proposes an adaptive weighting multimodal fusion classification system. The system uses handcrafted features as a supplement to abstract features, and accurately classifies steel surface defects in completely different feature representation spaces. Based on the NEU steel plate surface defect benchmark database, the classification results of feature sets before and after fusion are compared and analyzed. The experimental results show that the classification accuracy of the fusion system is improved by at least 3.4% compared with that before fusion, and the final accuracy rate is 99.0%, which proves the effectiveness of the proposed system.

Mud Ring Optimization Algorithm with Deep Learning Model for Disease Diagnosis on ECG Monitoring System.

Due to the tremendous growth of the Internet of Things (IoT), sensing technologies, and wearables, the quality of medical services has been enhanced, and it has shifted from standard medical-based health services to real time. Commonly, the sensors can be combined as numerous clinical devices to store the biosignals generated by the physiological actions of the human body. Meanwhile, a familiar method with a noninvasive and rapid biomedical electrocardiogram (ECG) signal can be used to diagnose and examine cardiovascular disease (CVD). As the growing number of patients is destroying the classification outcome because of major changes in the ECG signal patterns among numerous patients, computer-assisted automatic diagnostic tools are needed for ECG signal classification. Therefore, this study presents a mud ring optimization technique with a deep learning-based ECG signal classification (MROA-DLECGSC) technique. The presented MROA-DLECGSC approach recognizes the presence of heart disease using ECG signals. To accomplish this, the MROA-DLECGSC technique initially preprocessed the ECG signals to transform them into a uniform format. In addition, the Stacked Autoencoder Topographic Map (SAETM) approach was utilized for the classification of ECG signals to identify the presence of CVDs. Finally, the MROA was applied as a hyperparameter optimizer, which assisted in accomplishing enhanced performance. The experimental outcomes of the MROA-DLECGSC algorithm were tested on the benchmark database, and the results show the better performance of the MROA-DLECGSC methodology compared to other recent algorithms.

Automatic Breast Mass Lesion Detection in Mammogram Image

Mammography imaging is one of the most successful techniques for breast cancer screening and detecting breast lesions. Detection of the Region of Interest (ROI) (where the possible abnormalities could be present) is the backbone for the success of any Computer-Aided Detection or Diagnosis (CADx) system. In this paper, to assist the CADx system, one computational model is proposed to detect breast mass lesions from mammogram images. At the beginning of the process, pectoral muscles from the mammograms are removed as a pre-processing step. Then by applying an automatic thresholding scheme with the required image processing techniques, different regions of breast tissues are ranked to detect the possible suspected region to refine the further segmentation task. One seeded region growing approach is proposed with an automatic seed selection criterion to detect the suspected region to segment the ROI. The proposed model has very less user intervention as maximum of the parameters are computed automatically. To evaluate the performance of the proposed model, it is compared with four different methods with six different evaluation metrics viz., Jaccard & Dice co-efficient, relative error, segmentation accuracy, error and Fowlkes–Mallows index (FMI). On the proposed model, 57 mammogram images are tested, consisting of four different cases that are collected from the publicly available benchmark database. The qualitative and quantitative analyses are performed to evaluate the proposed model. The best dice co-efficient, Jaccard co-efficient, accuracy, error and FMI values observed are 0.9506, 0.9471, 95.62%, 4.38% and 0.932, respectively. The superiority of the model over six state-of-the-art compared methods is well evident from the experimental results.

Consistency Among the Results of the ASPIS Iron88, PCA Replica, and PCA ORNL Benchmark Experiments

Iron is an essential element in the construction materials for fission and fusion reactors. Due to its complexity, the evaluation of iron cross sections continues to represent a challenge for the international nuclear data community. A comprehensive validation of any new nuclear data evaluation (and the computational procedure) against experimental benchmarks is therefore needed. The shielding benchmark database SINBAD includes relatively numerous experiments with iron as a shielding material; altogether, 27 benchmarks and several more are known but have not yet been evaluated in the database. However, in order to use the benchmark information with confidence and to rely on the predictions based on integral benchmark calculations, it is crucial to verify the quality and accuracy of the measurements themselves, as well as the (completeness of) available experimental information. This is done in the scope of the benchmark evaluation process. A further check of the reliability of the experimental information can be achieved by intercomparing the results of similar types of benchmark experiments and checking the consistency among them.

A New Remote Hyperspectral Imaging System Embedded on an Unmanned Aquatic Drone for the Detection and Identification of Floating Plastic Litter Using Machine Learning

This paper presents a new Remote Hyperspectral Imaging System (RHIS) embedded on an Unmanned Aquatic Drone (UAD) for plastic detection and identification in coastal and freshwater environments. This original system, namely the Remotely Operated Vehicle of the University of Littoral Côte d’Opale (ROV-ULCO), works in a near-field of view, where the distance between the hyperspectral camera and the water surface is about 45 cm. In this paper, the new ROV-ULCO system with all its components is firstly presented. Then, a hyperspectral image database of plastic litter acquired with this system is described. This database contains hyperspectral data cubes of different plastic types and polymers corresponding to the most-common plastic litter items found in aquatic environments. An in situ spectral analysis was conducted from this benchmark database to characterize the hyperspectral reflectance of these items in order to identify the absorption feature wavelengths for each type of plastic. Finally, the ability of our original system RHIS to automatically recognize different types of plastic litter was assessed by applying different supervised machine learning methods on a set of representative image patches of marine litter. The obtained results highlighted the plastic litter classification capability with an overall accuracy close to 90%. This paper showed that the newly presented RHIS coupled with the UAD is a promising approach to identify plastic waste in aquatic environments.

Vehicle Make and Model Recognition as an Open-Set Recognition Problem and New Class Discovery

One of the main limitations of traditional neural-network-based classifiers is the assumption that all query data are well represented within their training set. Unfortunately, in real-life scenarios, this is often not the case, and unknown class data may appear during testing, which drastically weakens the robustness of the algorithms. For this type of problem, open-set recognition (OSR) proposes a new approach where it is assumed that the world knowledge of algorithms is incomplete, so they must be prepared to detect and reject objects of unknown classes. However, the goal of this approach does not include the detection of new classes hidden within the rejected instances, which would be beneficial to increase the model’s knowledge and classification capability, even after training. This paper proposes an OSR strategy with an extension for new class discovery aimed at vehicle make and model recognition. We use a neuroevolution technique and the contrastive loss function to design a domain-specific CNN that generates a consistent distribution of feature vectors belonging to the same class within the embedded space in terms of cosine similarity, maintaining this behavior in unknown classes, which serves as the main guide for a probabilistic model and a clustering algorithm to simultaneously detect objects of new classes and discover their classes. The results show that the presented strategy works effectively to address the VMMR problem as an OSR problem and furthermore is able to simultaneously recognize the new classes hidden within the rejected objects. OSR is focused on demonstrating its effectiveness with benchmark databases that are not domain-specific. VMMR is focused on improving its classification accuracy; however, since it is a real-world recognition problem, it should have strategies to deal with unknown data, which has not been extensively addressed and, to the best of our knowledge, has never been considered from an OSR perspective, so this work also contributes as a benchmark for future domain-specific OSR.

Handling similar looking disease symptoms in plants using dilation and feature reuse

For the last decade, the use of deep learning techniques in plant leaf disease recognition has seen a lot of success. Pretrained models and the networks trained from scratch have obtained near-ideal accuracy on various public and self-collected datasets. However, symptoms of many diseases found on various plants look similar, which still poses an open challenge. This work takes on the task of dealing with classes with similar symptoms by proposing a trained-from-scratch shallow and thin convolutional neural network employing dilated convolutions and feature reuse. The proposed architecture is only four layers deep with a maximum width of 48 features. The utility of the proposed work is twofold: (1) it is helpful for the automatic detection of plant leaf diseases and (2) it can be used as a virtual assistant for a field pathologist to distinguish among classes with similar symptoms. Since dealing with classes with similar-looking symptoms is not well studied, there is no benchmark database for this purpose. We prepared a dataset of 11 similar-looking classes and 5, 108 images for experimentation and have also made it publicly available. The results demonstrate that our proposed model outperforms other recent and state-of-the-art models in terms of the number of parameters, training & inference time, and classification accuracy.

Avoiding Optimal Mean Robust and Sparse BPCA with L1-norm Maximization

<p>Recently, the robust PCA/2DPCA methods have achieved great success in subspace learning. Nevertheless, most of them have a basic premise that the average of samples is zero and the optimal mean is the center of the data. Actually, this premise only applies to PCA/2DPCA methods based on L2-norm. The robust PCA/2DPCA method with L1-norm has an optimal mean deviate from zero, and the estimation of the optimal mean leads to an expensive calculation. Another shortcoming of PCA/2DPCA is that it does not pay enough attention to the instinct correlation within the part of data. To tackle these issues, we introduce the maximum variance of samples’ difference into Block principal component analysis (BPCA) and propose a robust method for avoiding the optimal mean to extract orthonormal features. BPCA, which is generalized from PCA and 2DPCA, is a general PCA/2DPCA framework specialized in part learning, can makes better use of the partial correlation. However, projection features without sparsity not only have higher computational complexity, but also lack semantic properties. We integrate the elastic network into avoiding optimal mean robust BPCA to perform sparse constraints on projection features. These two BPCA methods (non-sparse and sparse) make the presumption of zero-mean data unnecessary and avoid optimal mean calculation. Experiments on reference benchmark databases indicate the usefulness of the proposed two methods in image classification and image reconstruction.</p> <p> </p>

Graph-based learning of nonlinear physiological interactions for classification of emotions

Emotion recognition has been drawing the attention of researchers and practitioners in recent years. While various research studies show successful applications to recognize and distinguish emotions based on physiological responses using machine learning techniques, less research to date is focused on the network properties of physiological interactions under different emotional states. To this end, we propose a multi-modal graph learning framework to quantify the interactions among physiological systems and present representative networks associated with emotional states. More specifically, we introduce a novel information-theoretic-based time delay stability to quantify complex interactions between physiological modalities. We test our quantification approach on three publicly available benchmark databases for emotion recognition and demonstrate the comparative performances of measuring the interactions of physiological systems in response to emotional states. Finally, we present the visualization of multi-modal physiological network topology, which may be useful for emotional interpretations in practice.

Development of a cloud-assisted classification technique for the preservation of secure data storage in smart cities

Cloud computing is the most recent smart city advancement, made possible by the increasing volume of heterogeneous data produced by apps. More storage capacity and processing power are required to process this volume of data. Data analytics is used to examine various datasets, both structured and unstructured. Nonetheless, as the complexity of data in the healthcare and biomedical communities grows, obtaining more precise results from analyses of medical datasets presents a number of challenges. In the cloud environment, big data is abundant, necessitating proper classification that can be effectively divided using machine language. Machine learning is used to investigate algorithms for learning and data prediction. The Cleveland database is frequently used by machine learning researchers. Among the performance metrics used to compare the proposed and existing methodologies are execution time, defect detection rate, and accuracy. In this study, two supervised learning-based classifiers, SVM and Novel KNN, were proposed and used to analyses data from a benchmark database obtained from the UCI repository. Initially, intrusions were detected using the SVM classification method. The proposed study demonstrated how the novel KNN used for distance capacity outperformed previous studies. The accuracy of the results of both approaches is evaluated. The results show that the intrusion detection system (IDS) with a 98.98% accuracy rate produces the best results when using the suggested system.

Lesion size of early cerebral infarction on a Non-Contrast CT influences detection ability in Cascade Mask Region-Convolutional neural networks

BackgroundCurrently, convolutional neural network (CNN) methods for early non-contrast brain CT (NCCT) infarction detection have limited accuracy. We suspect this may be due to the significant scale variance of lesion size, which would result in small infarctions not being properly detected. Materials and MethodsPatients’ early NCCT infarction signs were labeled and divided into training/validation and test datasets. The 1st Cascade Mask Region-Convolutional Neural Network (Cascade Mask R-CNN) model (1st model) was train/validated. The test dataset, with added normal controls, was used to evaluate the performances. We further divided the lesions into large and small ones by statistics. We filtered the test dataset’s ‘small’ lesions and re-examined the 1st model. Finally, we filtered the ‘small’ lesions of the training/validation datasets to train/validate the 2nd Cascade Mask R-CNN (2nd model), and the performances were examined by the filtered test dataset. ResultsIn total, 266 ischemic stroke patients (Male/Female 163/103, age 69.9 ± 14.1 years) were included, and the duration from last known well to NCCT was 28.5 ± 39.3 h. Two hundred twenty patients were placed in training/validation datasets and 46 were assigned to the test dataset. Twenty normal control cases (Male/Female 14/6, age 68.7 ± 10.8 years) were added into the test dataset. The results were Sensitivity: 74.0%; Specificity: 91.8%; Accuracy: 86.1%; Precision: 80.9%; and F1-score: 77.3%. Based on the histogram, large lesions were > 2,012 pixels (4.8 cm2), and small lesions were ≤ 2,012 pixels. The 1st model’s performances by the filtered test dataset were Sensitivity: 96.9%; Specificity: 92.9%; Accuracy: 93.6%; Precision: 74.1%; and F1-score: 84.0%. The 2nd model’s performances were Sensitivity: 97.1%; Specificity: 96.4%; Accuracy: 96.5%; Precision: 85.7%; and F1-score: 91.0%. ConclusionsThe target objects in most benchmark databases do not have the property of significant scale variance, which means commonly developed AI models focus on the issue of translational variance. The conventional CNN-based AI models do have the property of translation equivariance. The impact of scale variance has rarely been explored in real applications. For patients with ischemic stroke, the lesion size may vary significantly. Experimental results showed that NCCT’s early infarction size can influence commonly used conventional CNN-based object detection and image segmentation modes, such as Cascade Mask R-CNN’s performance.

Pension fund investments in infrastructure

AbstractInfrastructure investment is increasingly important for institutional investors. This paper focuses on pension fund allocations to infrastructure, using the CEM Benchmarking database. We document that larger pension funds, public pension funds, and pension funds with a higher allocation to other alternative assets are more likely to invest in infrastructure. Pension funds across the globe face infrastructure investment costs that are comparable, but we observe significant scale advantages in infrastructure investments, with larger pension funds exhibiting lower investment costs, and higher net returns. Importantly, infrastructure has been among the best-performing asset classes, as measured by net returns, and we document persistence in pension fund infrastructure investment performance over a one-year horizon.

Ebola optimization with modified DarkNet‐53 model for scene classification and security on Internet of Things in smart cities

Remote sensing (RS) has developed significantly with the progress of the Internet of Things (IoT) which is allowed the cheap and fast acquisition of data in millions and billions of interrelated devices utilized throughout the whole world. RS scene classifier that purposes for classifying scene types for RS images has wide applications in several domains like urban planning, national defence security, environmental monitoring, and natural hazard detection. State-of-the-art deep learning (DL) successes are performed in a novel wave of RS scene classification applications, but it is the absence of explainability and trustworthiness. An intrusion detection system (IDS) plays a vital role to ensure security in the RS-based IoT environment. In this aspect, this study presents an ebola optimization algorithm with deep learning-based scene classification and intrusion detection (EOADL-SCID) technique on IoT-enabled remote sensing images. The aim of the EOADL-SCID system lies in the effectual scene classification of remote sensing images and intrusion detection. It involves a two-stage procedure. In the initial stage, the EOADL-SCID algorithm involves a modified DarkNet-53 feature extractor, EOA-based hyperparameter tuning, and graph convolution network (GCN) based classification. Next, in the second stage, the intrusion detection process takes place via two subprocesses namely variational autoencoder (VAE) based intrusion detection and skill optimization algorithm (SOA) based parameter tuning. The simulation outcomes of the EOADL-SCID approach are tested utilizing two benchmark databases and the experimental outcomes highlighted the improved performance of the EOADL-SCID algorithm on scene classification and intrusion classification processes.

Kernel-Induced Possibilistic Fuzzy Associate Background Subtraction for Video Scene

The background subtraction (BGS) technique is popularly used for many surveillance systems, segmenting the foreground by subtracting the modeled background from the image sequences. The effectiveness of any BGS technique depends on the robustness of the constructed background model. It is to be noted that many BGS schemes are affected by the inclusion of either noisy pixels in background construction or parameters of generative models. In this regard, we propound an idea of a kernel-induced possibilistic fuzzy associated BGS scheme for local change detection from a fixed camera captured sequence. The proposed scheme follows two stages: background training and foreground segmentation. In the background construction stage, each pixel is modeled using a possibilistic fuzzy cost function in kernel-induced space. The use of the induced kernel function will project the low-dimensional data into a higher dimensional space and the use of the possibilistic function will construct a robust background model based on the density of the data in the temporal domain avoiding the noisy and outlier points. The performance of the proposed scheme is tested on three benchmark databases. The effectiveness of the proposed scheme is evaluated on different performance evaluation measures: precision, recall, F-measure, and average similarity. We corroborate our findings by comparing them against 19 state-of-the-art existing BGS techniques.

Viewport-Sphere-Branch Network for Blind Quality Assessment of Stitched 360° Omnidirectional Images

Compared with conventional images/videos, omni-directional data records rich information with higher resolution and wider Field-of-View. Moreover, the stitching distortions introduced in the panoramic content generation process make the quality assessment task more challenging. Targeting at designing an accurate and fast stitched 360° omnidirectional image quality evaluator, we propose a Viewport-Sphere-Branch Network (VSBNet) via dual-branch quality estimation. Specifically, for the viewport quality estimation, we extract distorted viewports around the stitching seams and conduct distortion rectification through a progressively complementary network to obtain pseudo-reference viewports. The qualitative and quantitative experiments validate that pseudo-reference viewports are reliable. Then, the differences between distorted and pseudo-reference viewports are quantified through transformer architecture to obtain quality scores of viewports. The introduction of pseudo-reference viewports can effectively improve the performance of the viewport quality prediction branch. To establish general scenario awareness and accurately evaluate the immersive experience, we extract feature representation through deformable convolutions to eliminate 2D-to-Sphere intrinsic sampling distortions and use multilayer perceptron to predict score of the whole sphere. The final prediction score is obtained by aggregating the quality scores from viewport and sphere branches. We evaluate the proposed VSBNet on two benchmark databases and results demonstrate that the combination of two branches can obtain more accurate results. Overall, our method is superior to existing full reference and no reference models designed for conventional images and 360° omnidirectional images.

A Case for Graphics-Driven Query Processing

Over the past decade, the database research community has directed considerable attention towards harnessing the power of GPUs in query processing engines. The proposed techniques have primarily focused on devising customized low-level mechanisms that utilize the raw hardware parallelism provided abundantly by GPU compute kernels. In this paper, we advocate a radically different approach - instead of dealing directly with hardware idiosyncrasies, to leverage the well-established graphics pipeline architecture baked into the GPU hardware. A variety of advantages accrue from this high-level abstraction: (a) Extracting the power of GPUs is outsourced to highly-optimized graphics drivers, thereby providing hardware-consciousness for free; (b) Query processing becomes agnostic to changes in GPU architectures (e.g. integrated vs discrete) and vendors, requiring only a change of drivers; (c) Contemporary graphics APIs also support a compute element, facilitating query operator designs that seamlessly straddle the compute and graphics worlds. As a proof of concept of the above vision, we implement here the workhorse Join and GroupBy operators using core graphics primitives. These implementations, based on the Vulkan API, have been evaluated over large benchmark databases on vanilla hybrid computing platforms. The experimental results indicate both substantive performance benefits (typically, around 2X faster) over existing approaches, as well as auto-tuned portability to new hardware platforms.

EDGE DETECTION TECHNIQUE BASED ON BILATERAL FILTERING AND ITERATIVE THRESHOLD SELECTION ALGORITHM AND TRANSFER LEARNING FOR TRAFFIC SIGN RECOGNITION

The traffic sign identification and recognition system (TSIRS) is an essential component for autonomous vehicles to succeed. The TSIRS helps to collect and provide helpful information for autonomous driving systems. The information may include limits on speed, directions for driving, signs to stop or lower the speed, and many more essential things for safe driving. Recently, incidents have been reported regarding autonomous vehicle crashes due to traffic sign identification and recognition system failures. The TSIRS fails to recognize the traffic signs in challenging conditions such as skewed signboards, scratches on traffic symbols, discontinuous or damaged traffic symbols, etc. These challenging conditions are presented for various reasons, such as accidents, storms, artificial damage, etc. Such traffic signs contain an ample amount of noise, because of which traffic sign identification and recognition become a challenging task for automated TSIRS systems. The proposed method in this paper addresses these challenges. The sign edge is a helpful feature for the recognition of traffic signs. A novel traffic sign edge detection algorithm is introduced based on bilateral filtering with adaptive thresholding and varying aperture size that effectively detects the edges from such noisy images. The proposed edge detection algorithm and transfer learning is used to train the Convolutional Neural Network (CNN) models and recognize the traffic signs. The performance of the proposed method is evaluated and compared with existing edge detection methods. The results show that the proposed algorithm achieves optimal Mean Square Error (MSE) and Root Mean Square Error (RMSE) error rates and has a better Signal to Noise Ratio (SNR) and Peak Signal to Noise Ratio (PSNR) ratio than the traditional edge detection algorithms. Furthermore, the precision rate, recall rate, and F1 scores are evaluated for the CNN models. With the German Traffic Sign Benchmark database (GTSRB), the proposed algorithm and Inception V3 CNN model gives promising results when it receives the edge-detected images for training and testing.

Analyzing Human Observer Ability in Morphing Attack Detection—Where Do We Stand?

Morphing attacks are based on the technique of digitally fusing two (or more) face images into one, with the final visage resembling the contributing faces. Morphed images not only pose a challenge to Face-Recognition Systems (FRS) but also challenge experienced human observers due to high quality, postprocessing eliminating any visible artifacts, and further the printing and scanning process. Few studies have concentrated on examining how human observers can recognize morphing attacks, even as several publications have examined the susceptibility of automated FRS to morphing attacks and offered morphing attack detection (MAD) approaches. MAD approaches base their decisions either on a single image with no reference to compare against (Single-Image MAD (S-MAD)) or using a reference image (Differential MAD (D-MAD)). One prevalent misconception is that an examiner’s or observer’s capacity for facial morph detection depends on their subject expertise, experience, and familiarity with the issue. No works have reported the specific results of observers who regularly verify identity (ID) documents for their jobs. As human observers are involved in checking ID documents having facial images, a lapse in their competence can result in significant societal challenges. To assess the observers’ proficiency, this research first builds a new benchmark database of realistic morphing attacks from 48 different subjects, resulting in 400 morphed images. Unlike the previous works, we also capture images from Automated Border Control (ABC) gates to mimic realistic border-crossing scenarios in the D-MAD setting with 400 probe images, to study the ability of human observers to detect morphed images. A new dataset of 180 morphing images is also produced to research human capacity in the S-MAD environment. In addition to creating a new evaluation platform to conduct S-MAD and D-MAD analysis, the study employs 469 observers for D-MAD and 410 observers for S-MAD who are primarily governmental employees from more than 40 countries, along with 103 control group members who are not examiners. The analysis offers intriguing insights and highlights the lack of expertise and failure to recognize a sizable number of morphing attacks by experienced observers. Human observers tend to detect morphed images to a lower accuracy as compared to the four automated MAD algorithms evaluated in this work. The results of this study are intended to aid in the development of training programs that will prevent security failures while determining whether an image is bona fide or altered.

Experimental assessment of the hydrodynamic characteristics of a bulk carrier in off-design conditions

This paper presents an extensive experimental study on the resistance and propulsion characteristics of the Japan Bulk Carrier in off-design conditions. The experimental campaign is carried out for a ship model at a scale of 40.264 for three loading conditions. Resistance, open water, and self-propulsion tests were performed at the Brodarski Institute for a wide range of speeds. Four extrapolation methods, based on different assumptions, were applied to predict the resistance and propulsion characteristics of the full-scale ship from model-scale experiments. The differences between the results obtained by different extrapolation methods and the effect of loading conditions on ship resistance and propulsion characteristics are discussed. This study provides a benchmark database for the validation of Computational Fluid Dynamics simulations.

Deep Learning-Based Anomaly Detection in Video Surveillance: A Survey.

Anomaly detection in video surveillance is a highly developed subject that is attracting increased attention from the research community. There is great demand for intelligent systems with the capacity to automatically detect anomalous events in streaming videos. Due to this, a wide variety of approaches have been proposed to build an effective model that would ensure public security. There has been a variety of surveys of anomaly detection, such as of network anomaly detection, financial fraud detection, human behavioral analysis, and many more. Deep learning has been successfully applied to many aspects of computer vision. In particular, the strong growth of generative models means that these are the main techniques used in the proposed methods. This paper aims to provide a comprehensive review of the deep learning-based techniques used in the field of video anomaly detection. Specifically, deep learning-based approaches have been categorized into different methods by their objectives and learning metrics. Additionally, preprocessing and feature engineering techniques are discussed thoroughly for the vision-based domain. This paper also describes the benchmark databases used in training and detecting abnormal human behavior. Finally, the common challenges in video surveillance are discussed, to offer some possible solutions and directions for future research.