Low False Alarm Rate Research Articles

ST-YOLOA: a Swin-transformer-based YOLO model with an attention mechanism for SAR ship detection under complex background.

A synthetic aperture radar (SAR) image is crucial for ship detection in computer vision. Due to the background clutter, pose variations, and scale changes, it is a challenge to construct a SAR ship detection model with low false-alarm rates and high accuracy. Therefore, this paper proposes a novel SAR ship detection model called ST-YOLOA. First, the Swin Transformer network architecture and coordinate attention (CA) model are embedded in the STCNet backbone network to enhance the feature extraction performance and capture global information. Second, we used the PANet path aggregation network with a residual structure to construct the feature pyramid to increase global feature extraction capability. Next, to cope with the local interference and semantic information loss problems, a novel up/down-sampling method is proposed. Finally, the decoupled detection head is used to achieve the predicted output of the target position and the boundary box to improve convergence speed and detection accuracy. To demonstrate the efficiency of the proposed method, we have constructed three SAR ship detection datasets: a norm test set (NTS), a complex test set (CTS), and a merged test set (MTS). The experimental results show that our ST-YOLOA achieved an accuracy of 97.37%, 75.69%, and 88.50% on the three datasets, respectively, superior to the effects of other state-of-the-art methods. Our ST-YOLOA performs favorably in complex scenarios, and the accuracy is 4.83% higher than YOLOX on the CTS. Moreover, ST-YOLOA achieves real-time detection with a speed of 21.4 FPS.

Modulation-Agnostic Spectrum Sensing based on Anomaly detection for Cognitive Radio

This work proposes a Spectrum Sensing (SS) scheme based on a Convolutional Autoencoder (CAE) for application in Cognitive Radio Networks. The channel occupancy is modeled as an anomaly detection problem. The CAE is trained only with noise-signal samples so that any random modulated signal observed in the wireless channel will be regarded as an outlier. The relationship between the detection threshold and the system performance is evaluated. Experiments demonstrate that the SS proposal can identify primary signal presence with better accuracy (high detection and low false alarm rate) than conventional Energy Detection.

Detection of infrared small target based on background subtraction local contrast measure and Gaussian structural similarity

Infrared (IR) small target detection, especially in a complex background, continues to present challenges in the low false alarm rate and high robustness. In this paper, a background subtraction local contrast measure (BSLCM) and Gaussian structural similarity (GSS) integrated structural model is proposed to detect IR small target. In the proposed model, BSLCM is used to suppress the complex background and enhance the target. GSS calculation is conducted to eliminate the high-brightened background residual and noise further. To evaluate the performance of the proposed method, real IR sequences and seven state-of-the-art (SOTA) methods were adopted. The results demonstrated that the BSLCM can suppress all types of strong background clutter and enhance the true target effectively.

A Mixed Intrusion Detection System utilizing K-means and Extreme Gradient Boosting

The intrusion detection system (IDS) plays an important part because it offers an efficient way to prevent and mitigate cyber attacks. Numerous deep learning methods for intrusion anomaly detection have been developed as a result of recent advances in artificial intelligence (AI) in order to strengthen internet security. The balance among the high detection rate (DR), the low false alarm rate (FAR) and disaster of dimensionality is the crucial apprehension while devising an effective IDS. For the binary classification of intrusion detection systems, we present in this study a mixed model called K-means-XGBoost consisting of K-means and (Extreme Gradient Boosting, XGBoost) algorithms. The distributed computation of our method is achieved in Spark platform to rapidly separate normal events and anomaly events. In phrases of accuracy, DR, F1-score, recall, precision, and error indices FAR, the proposed model’s performance is measured via the well-known dataset of NSL-KDD. The experimental outcomes indicate that our method is outstandingly better among accuracy, DR, F1-score, training time, and processing speed, compared to other models which are recently created. In particular, the accuracy, F1-score, and DR of the proposed model can achieve as high as 93.28%, 94.39%, and 99.22% in the NSL-KDD dataset, respectively.

Ensemble of supervised and unsupervised deep neural networks for stock price manipulation detection

Illegal practices that cause stock prices to vary from their fair values are known as stock price manipulations. Our prior study used unsupervised deep learning to detect these unlawful activities, in which the autoencoder model could catch five out of six cases in the Stock Exchange of Thailand (SET). Trading actions of good governance equities in a limit order book format were used to train the model, and abnormalities were viewed as manipulation. This unsupervised technique has the strength of discovering never-before-seen patterns, although it has a low recall rate. In this work, the findings were improved by incorporating knowledge about the popular pump-and-dump pattern into another supervised deep learning model, which has a strength in catching known patterns. To combine the advantages of both models, an ensemble of supervised and unsupervised deep neural networks was proposed. The main contribution of this paper is to demonstrate that the new ensemble model achieved a higher detection rate than the unsupervised model alone. The supervised learning model used was a long short-term memory (LSTM) network, the unsupervised learning model used was an LSTM-based autoencoder, and the stack classifier in the ensemble model was a support vector machine (SVM). The proposed method was evaluated on six real manipulation cases from the SET. The empirical results showed that our ensemble model could improve the detection strength, achieved high accuracy on pump-and-dump unseen data, and could identify six out of six real manipulation cases with low false alarm rates.

Graph Spatiotemporal Pattern Learning Network for Real-Time Road Network Traffic Abnormal Incident Detection

To improve the efficiency of detecting abnormal traffic incidents on the road network and reduce the false alarm rate, a real-time traffic anomaly detection framework based on a graph spatiotemporal pattern learning (GSTPL) network is proposed. In this framework, a traffic pattern search algorithm based on a fluctuation similarity measure is designed to screen traffic flow data with the same traffic pattern, and a traffic pattern graph tuple is constructed as the input of the network model to avoid the sample imbalance problem and the effect of single-sample randomness for traffic pattern learning. Then the GSTPL network is designed to extract, unsupervised, the traffic spatiotemporal pattern features and make a reasonable prediction of future traffic parameters as the basis for anomaly evaluation. To further restrain the effect of random fluctuations in traffic flow parameters, an abnormal state evaluation method is designed to calculate the anomaly state likelihood by prediction error distribution learning. The overall detection framework realizes stable prediction of network key node traffic parameters by using spatiotemporal pattern features to construct the traffic pattern graph tuple, and gives incident evaluation results in real time by combination with the detection data. The experiment uses I90 and I405 highway traffic data in Seattle, WA, from 2015. Through comparative analysis, the proposed incident detection method based on GSTPL has a higher detection rate and lower false alarm rate, can adaptively learn dynamic changes of the traffic pattern, and has strong adaptability and stability to different traffic environments.

Validation of INSAT-3D/3DR based nowcasting rain occurrences for heavy rainfall using Hydro-Estimator product

Satellite-based Nowcasting methods are of utmost importance, especially in the Indian monsoonal region, which experiencing heterogeneity in rainfall structures. In this context, the INSAT-3D/3DRsatellite-based model for Nowcasting of Extreme orographic Rain occurrences (NETRA) for the Western Himalayan region was developed, and was later implemented for the entire Indian landmass, provides near real-time alerts for heavy rainfall events through the web portal https://www.mosdac.gov.in with an update frequency of half an hour. The societal application of this model is tightly linked with its validation over different seasons and years. In the present paper, we have validated two years of rainfall alarm provided by the model with the help of Quantitative Precipitation Estimate (QPE) using the INSAT-3D/3DR Hydro-Estimator (H-E) product. Regarding the frequency of rainfall occurrences, INSAT-3D/3DR satellite rainfall products excelled in capturing the rainfall pattern both spatially and temporally. While the QPE correlation is 0.1, for heavy rainfall events, H-E demonstrates better skill and correlation (r > 0.7) in detecting heavy rainfall with an accuracy of 20 mm and good pattern matching with actualrainfall. Itis observed that for the months of May to October, the probability of detection is quite high (∼95%) with a low false alarm rate. In case of extreme events also, the algorithm performs quite well as is shown by extreme dependency indices like the (Extreme Dependency Score)EDS, (Extreme Dependency Index) EDI, and (Symmetric Extreme Dependency Index) SEDI. For the Indian Monsoonal region, which experiences a significant loss of life and property due to heavy rainfall events, this satellite-based nowcasting alert system may have a substantial societal impact.

Geostationary active fire products validation: GOES-17 ABI, GOES-16 ABI, and Himawari AHI

ABSTRACT The current generation of geostationary Earth-observing satellites provide spectral bandpass, spatial resolution and imaging frequency characteristics well suited to near-continuous active fire detection and monitoring. The earliest of these systems-SEVIRI on-board EUMETSAT’s MSG series-has operated since 2004, and more recently the capability has been expanded globally with the ABI on-board NOAA’s GOES-16 and GOES-17 satellites, and the AHI on-board JMA’s Himawari-8 and Himawari-9. At present, the NOAA and EUMETSAT operational geostationary active fire products are available based on two different algorithms: the Fire Detection and Characterization (FDC) product operating with data from GOES-16 and −17, and FRP-PIXEL active fire products from GOES, Himawari and MSG. We have conducted a comprehensive accuracy assessment of these geostationary fire products across two seasons (1 January–31 March 2020 and 1 July–30 September 2020), based on comparison to Landsat active fire detections made simultaneously (±5 minutes of geostationary overpass time) with the geostationary data. Compared to Landsat we find (i) low false alarm rates, ranging between 4%–7% (FDC) and 2%-6% (FRP-PIXEL)- depending on the season and hemispheric-disk for high confidence pixels, (ii) a reduction in this false alarm rate for FDC due to algorithm changes made since our prior (2018) validation effort (48% false alarms in summer 2018 compared to 4% in summer 2020 for high confidence pixels), and (iii) comparable active fire pixel detection rates for the FDC product (high confidence fire pixel classes only) and the matching FRP-PIXEL product (all fire pixel confidence classes). Overall, the performance of these geostationary products is shown to be strong and complementary in that the FRP-PIXEL product has fewer false alarms but a lower detection rate, whereas the FDC product detects more fire pixels but with a much higher false alarm rate.

Hybrid intelligent intrusion detection system for multiple Wi‐Fi attacks in wireless networks using stacked restricted Boltzmann machine and deep belief networks

SummaryThe recent cutting edge mobile and internet advancements have increased Internet of Things based gadgets more pervasive and cohesive into our day to day lives. There emerges a need to beat the security challenges, prevalently came about because of the uncovered idea of a remote medium, for example a Wi‐Fi network setup. The linked gadgets are ubiquitous, creating high‐dimensional information on an expansive scale. It tends to be seen those illegitimate actions like illegal information access, information stealing, information alteration different other interruption exercises are quickly developing during a decade ago. Wireless network data transfers will be anticipated to surge strongly in the coming years. Feature learning, still, can evade the probable complications, possibly formed by means of large wireless network dimensions. Therefore, the proposed system focuses on developing a Hybrid Intelligent Intrusion Detection System (IDS) for detecting multiple assaults such as Impersonation, Injection, flooding along with other unknown malicious Wi‐Fi attacks using Deep Learning and Machine Learning algorithms. The proposed system is investigated on well referenced Aegean Wi‐Fi Intrusion Dataset (AWID) and the experimental results prove that the utility of deep learning algorithms in feature representation and classification has achieved higher detection accuracy along with lower false alarm rate. To get an optimum feature set and perform dimensionality reduction on Wi‐Fi network data, the proposed framework uses Deep Belief Network (DBN) with stacked layers of Restricted Boltzmann Machine (RBM). In order to identify and categories various Wi‐Fi threats, Deep Neural Networks, and algorithms like SVM, MLKNN, and RAkEL are employed. The anticipated DBN model was given an optimization strategy to enhance classification performance. The optimization strategy involved selecting the best possible feature subset using a feature selection technique, optimizing the DBN model using the best possible hyperparameter values, less reconstruction error values, and utilizing the regularization method to account for weight decay. The experimental findings demonstrate that the use of deep learning algorithms in feature representation and classification has produced improved detection accuracy coupled with a decreased false alarm rate when applied to the proposed hybrid scheme based Intelligent IDS using the well‐referenced AWID dataset.

ProTopormer: Toward Understandable Fault Diagnosis Combining Process Topology for Chemical Processes

In modern chemical processes, fault detection and diagnosis (FDD) is a key part of abnormal situation management (ASM). As researchers continue to improve the fault diagnosis performance of different models, the emphasis on model interpretability and explainability studies has increased in recent years. In this paper, a novel model, ProTopormer, was proposed for fault diagnosis of chemical processes. Self-attention mechanism and process topology knowledge were fully combined to achieve high model performance and good interpretability. Experiments on Tennessee Eastman process showed that the model achieved a high diagnosis rate and a low false alarm rate. Attention weights were visualized and quantitatively analyzed to identify key variables for fault diagnosis, which showed strong explanations for the root causes of different faults.

Data mining based network intrusion detection method in the environment of IoT

AbstractWith the rapid development of Internet of Things (IoT) and Internet technology, the network is becoming more and more important and strict. The IoT networks that access to a large number of devices make network intrusion more convenient. However, because the traditional network intrusion detection methods have the problems of low detection rate and high false alarm rate, this paper proposes a network intrusion detection method based on data mining. Firstly, a pre‐processing model is established to process the initial data. Then, according to the characteristics of network behavior, a new feature subset selection process is designed. Based on the above data mining and analysis process, a dynamic detection generation rule matching with data mining is set to detect abnormal network intrusion traffic. Then, the network intrusion detection based on data mining is completed. The experimental results show that, compared with other methods, the network intrusion detection method based on data mining has higher network intrusion detection rate and lower false alarm rate, which is helpful to improve the detection accuracy.

Dim and Small Space-Target Detection and Centroid Positioning Based on Motion Feature Learning

The detection of dim and small space-targets is crucial in space situational awareness missions; however, low signal-to-noise ratio (SNR) targets and complex backgrounds pose significant challenges to such detection. This paper proposes a space-target detection framework comprising a space-target detection network and a k-means clustering target centroid positioning method. The space-target detection network performs a three-dimensional convolution of an input star image sequence to learn the motion features of the target, reduces the interference of noise using a soft thresholding module, and outputs the target detection result after positioning via the offsetting branch. The k-means centroid positioning method enables further high-precision subpixel-level centroid positioning of the detection network output. Experiments were conducted using simulated data containing various dim and small space-targets, multiple noises, and complex backgrounds; semi-real data with simulated space-targets added to the real star image; and fully real data. Experiments on the simulated data demonstrate the superior detection performance of the proposed method for multiple SNR conditions (particularly with very low false alarm rates), robustness regarding targets of varying numbers and speeds, and complex backgrounds (such as those containing stray light and slow motion). Experiments performed with semi-real and real data both demonstrate the excellent detection performance of the proposed method and its generalization capability.

Track Detection of Underwater Moving Targets Based on CFAR

In this paper, we propose a 2D-Weibull-Constant False Alarm Rate (2D-Weibull-CFAR) detection algorithm to solve the problem that detecting current underwater targets is difficult due to the influence of reverberation noise. Specifically, referring to the idea that CFAR uses the probability distribution of reference units to detect objects, this paper introduces the pixel distribution of reverberation noise into the CFAR detector. After that, the probability distribution of the extracted reference units is estimated, and then the adaptive detection threshold is obtained to achieve reliable detection of underwater targets. Finally, the Hough transform extracts the trajectory of the detection results. The experimental test shows that the algorithm in this paper can solve the problem of false alarms and missed alarms in detecting targets hidden in the reverberation noise. The algorithm in this paper can effectively detect the target in the reverberation noise. The detection results show that the algorithm in this paper has higher accuracy and lower false alarm rate than the comparison algorithm.

Condition Monitoring-Oriented Wind Turbine Early Fault Rule K-Nearest Neighbor Matching Method

Due to the lack of data cleaning links in traditional methods, there are many false alarms and the problem of susceptibility to noise during the monitoring process. For this reason, this paper proposes a condition monitoring-oriented wind turbine early fault rule k nearest neighbor matching method. Obtain static and dynamic parameters through identification of wind turbine operating state parameters, use nuclear density estimation repair method to repair missing data, use rule k nearest neighbor matching method to remove abnormal data and noise data, and use ReliefF algorithm to screen wind turbine operating faults based on data processing results feature. Finally, the uncertainty of the fault status of the wind turbine is analyzed, and the early fault monitoring platform of the wind turbine is established based on the analysis result to realize the early fault monitoring of the wind turbine. Experimental results show that the method has better anti-noise performance and lower fault false alarm rate.

A Honeybee-Inspired Framework for a Smart City Free of Internet Scams.

Internet scams are fraudulent attempts aim to lure computer users to reveal their credentials or redirect their connections to spoofed webpages rather than the actual ones. Users' confidential information, such as usernames, passwords, and financial account numbers, is the main target of these fraudulent attempts. Internet scammers often use phishing attacks, which have no boundaries, since they could exceed hijacking conventional cyber ecosystems to hack intelligent systems, which emerged recently for the use within smart cities. This paper therefore develops a real-time framework inspired by the honeybee defense mechanism in nature for filtering phishing website attacks in smart cities. In particular, the proposed framework filters phishing websites through three main phases of investigation: PhishTank-Match (PM), Undesirable-Absent (UA), and Desirable-Present (DP) investigation phases. The PM phase is used at first in order to check whether the requested URL is listed in the blacklist of the PhishTank database. On the other hand, the UA phase is used for investigation and checking for the absence of undesirable symbols in uniform resource locators (URLs) of the requested website. Finally, the DP phase is used as another level of investigation in order to check for the presence of the requested URL in the desirable whitelist. The obtained results show that the proposed framework is deployable and capable of filtering various types of phishing website by maintaining a low rate of false alarms.

Detection of Abnormal Data in GNSS Coordinate Series Based on an Improved Cumulative Sum

The global navigation satellite system (GNSS), as a high-time resolution and high-precision measurement technology, has been widely used in the field of deformation monitoring. Owing to the influence of uncontrollable factors, there are inevitably some abnormal data in the GNSS monitoring series. Thus, it is necessary to detect and identify abnormal data in the GNSS monitoring series to improve the accuracy and reliability of the deformation disaster law analysis and warning. Many methods can be used to detect abnormal data, among which the statistical process control theory, represented by the cumulative sum (CUSUM), is widely used. CUSUM usually constructs statistics and determines control limits based on the threshold criteria of the average run length (ARL) and then uses the control limits to identify abnormal data in CUSUM statistics. However, different degrees of the ‘trailing’ phenomenon exist in the interval of abnormal data identified by the algorithm, leading to a higher false alarm rate. Therefore, we propose an improved CUSUM method that uses breaks for additive season and trend (BFAST) instead of ARL-based control limits to identify abnormal data in CUSUM statistics to improve the accuracy of identification. The improved CUSUM method is used to detect abnormal data in the GNSS coordinate series. The results show that compared with CUSUM, the improved CUSUM method shows stronger robustness, more accurate detection of abnormal data, and a significantly lower false alarm rate.

Unsupervised Deep Tensor Multitask Anomaly Detection With Rule Adaptation for Online Early Fault Evaluation

This paper tries to solve the problem of online early fault detection (EFD) with unlabeled streaming data by addressing the following challenges: 1) no state information is available for online data; 2) online working condition is not determined in advance; 3) False alarm should be avoided. This paper first proposes a deep tensor multi-task anomaly detection model (Tensor-MAD) with rule adaptation for the online EFD. Running on multi-task learning architecture, Tensor-MAD builds a new tensorized pooling filter to keep the essential information of each task’s operating status from noisy data. With hypersphere-based one-class detection rule representation, Tensor-MAD further constructs a new rule adaptation mechanism to transfer the detection rule from offline labeled data to the online unlabeled data. A training algorithm with an alternating minimization scheme is also provided to update tensor decomposition and rule adaptation. Then the optimal information filter level and the rule adaptation degree can be determined. Based on the obtained anomalies, this paper proposes a non-parametric alarm threshold setting method based on the sequential accumulation of anomaly probability. This threshold can be adaptively chosen once an expected false alarm rate is given. A rationality proof is also provided. Experimental results on the IEEE PHM Challenge 2012 bearing dataset demonstrate that the proposed approach can adaptively and accurately evaluate the early fault occurrence from unlabeled streaming data. More importantly, the proposed approach has a much lower false alarm rate and a faster convergence speed, providing an easy-to-deploy and reliable solution for online EFD.

A trend-based method for the prediction of offshore wind power ramp

Wind power ramp is a destructive event accompanied with a sharp change of wind power. Offshore wind power is expected to receive more attention for it can harvest consistent and strong winds. In this paper, a ramp detection framework based on the swinging door algorithm (SDA) and the mergence of correct ramps is proposed. Meanwhile, a trend-based prediction method (T-Method) is proposed to predict offshore wind power ramps. The ramp detection framework is utilized to identify wind power ramp events (WPREs) and label original data according to the detection results. After that, the labeled data is selected as the input of recurrent neural network to produce wind power prediction results. Finally, the prediction results are detected by the ramp detection framework to produce WPREs prediction results. Four recurrent neural network models and two traditional methods are applied to two offshore wind power datasets for corroborating the effectiveness of our method. Comparative experiments show that our proposed method performs excellent in all evaluation indicators. The proposed WPREs prediction method improves the performance of WPREs prediction under the premise of low false alarm rate and missing alarm rate. It provides early warning for power system operators to reduce the harm of WPREs.

Fire Warning Algorithm based on Combining R-S Fusion Layer Images with Convolutional Neural Networks

In order to overcome the problems of low detection rate and high false alarm rate of fire detection system, a video fire detection algorithm based on R-S weighted model and convolutional neural network is proposed. First, the R-S fusion layer image is obtained by RGB color space and HSI color space; Then, the convolutional neural network is used to construct the fire detection algorithm. During the training process, the extracted R-S fusion layer image is sent to the convolutional neural network for training. During the detection process, the test image is sent to the model for detection. The experimental results show that the method has more effective flame detection effect.

An intelligent DDoS attack detection tree-based model using Gini index feature selection method

Cyber security has recently garnered enormous attention due to the popularity of the Internet of Things (IoT), intelligent devices’ rapid growth, and a vast number of real-life applications. As a result, detecting threats and constructing an efficient Intrusion detection system (IDS) have become crucial in today's security requirements. Withal, the large amount of high dimensional data might influence detection effectiveness and raise the computation requirements. Artificial Intelligence (AI) has recently attracted much attention and is widely used to build intelligent IDSs to preserve data confidentiality, integrity, and availability. Distributed denial of service (DDoS) is a denial of service (DoS) variant mainly targeting asset availability. Preventing DoS at the network or infrastructure level typically depends on implementing an IDS. This paper proposes a novel intelligent DDoS attack detection model based on a Decision Tee (DT) algorithm and an enhanced Gini index feature selection method. Our approach is evaluated on the UNSW-NB15 dataset, which contains 1,140,045 samples and is more recent and comprehensive than those used in previous works. Our system achieved an overall accuracy of 98%, outperforming baseline models that used more advanced algorithms such as Random Forest and XGBoost. Our enhanced Gini index feature selection method allowed us to select only 13 out of 45 security features, significantly reducing the data dimensionality and avoiding overfitting issues. Our model also has a lower false alarm rate, misclassifying only 2% of the testing instances. Our approach is, therefore, highly effective and efficient, with the potential to be used in real-world network security applications.