High False Alarm Rate Research Articles

Process Monitoring Using a Novel Robust PCA Scheme

Outliers may cause model deviation and then affect the monitoring performance and hence it is a challenging problem for process monitoring. The robust principal component analysis (RPCA) approach, which describes outlier components with a sparse matrix and identifies these components using the sparse matrix recovery approach, is the most commonly used method to solve the model deviation problems caused by outliers. However, because the existing mathematical tools can only obtain a nonsparse matrix with small element values, RPCA performs poorly during process monitoring. In this paper, we propose a novel robust PCA scheme called moment-based RPCA (MRPCA). In the offline training stage, MRPCA adopts a novel outlier selection mechanism based on the difference between the higher-order and second-order central moments to select outlier samples; in the online monitoring stage, MRPCA adopts an outlier detection mechanism to distinguish outliers from fault data. Using the aforementioned mechanisms, MRPCA achieves high fault detection and low false alarm rates in tests of a numerical model and the Tennessee Eastman process.

Sensor Selection Embedded in Active Fault Diagnosis Algorithms

A methodology is presented for simultaneous sensor selection and design of fault diagnosis tests in complex systems, for which steady-state or dynamic models are available. The method assesses all possible sensor combinations for their information with respect to system faults in the presence of uncertainty and chooses sensors based on their contribution to information gain. In particular, sensors are cast as binary variables in the normalized Fisher information matrix, which is used to calculate the optimal fault detection and isolation (FDI) test designs by manipulating admissible system inputs. Test design optimization is always naturally embedded in the proposed method. Then, the Kullback–Leibler divergence and the Hellinger distance are used to explore the isolation capability of an FDI test when uncertainty is considered for the system inputs and parameters. The FDI tests are deployed and verified using a principal component analysis and the $k$ -nearest neighbor classification algorithm. The capability of the method to detect and isolate faults at high correct classification rates and low false alarm rates is demonstrated. The tool-chain proposed is tested on a benchmark three-tank system at various levels of measurement noise and uncertainty. The benefits and drawbacks of each step of the proposed method are assessed and discussed along with their computational footprint.

Forest Fire Smoke Recognition Based on Anchor Box Adaptive Generation Method

There are major problems in the field of image-based forest fire smoke detection, including the low recognition rate caused by the changeable and complex state of smoke in the forest environment and the high false alarm rate caused by various interferential objects in the recognition process. Here, a forest fire smoke identification method based on the integration of environmental information is proposed. The model uses (1) the Faster R-CNN as the basic framework, (2) a component perception module to generate a receptive field of integrated environmental information through separable convolution to improve recognition accuracy, and (3) a multi-level Region of Interest (ROI)pooling structure to reduce the deviation caused by rounding in the ROI pooling process. The results showed that the model achieved a recognition accuracy rate of 96.72%, an Intersection Over Union (IOU) of 78.96%, and an average recognition speed for each picture of 1.5 ms; the false alarm rate was 2.35% and the false-negative rate was 3.28%. Compared with other models, the proposed model can effectively enhance the recognition accuracy and recognition speed of forest fire smoke, which provides a technical basis for the real-time and accurate detection of forest fires.

Intense L-Band Solar Radio Bursts Detection Based on GNSS Carrier-To-Noise Ratio Decrease over Multi-Satellite and Multi-Station.

Intense solar radio bursts (SRBs) can increase the energy noise and positioning error of the bandwidth of global navigation satellite system (GNSS). The study of the interference from intense L-band SRBs is of great importance to the steady operation of GNSS receivers. Based on the fact that intense L-band SRBs lead to a decrease in the carrier-to-noise ratio () of multiple GNSS satellites over a large area of the sunlit hemisphere, an intense L-band SRB detection method without the aid of a radio telescope is proposed. Firstly, the valley period of a single satellite at a single monitoring station is detected. Then, the detection of SRBs is achieved by calculating the intersection of multiple satellites and multiple stations. The experimental results indicate that the detection rates of GPS L2 and GLONASS G2 are better than those of GPS L1 L5, GLONASS G1, and Galileo E1 E5. The detection rate of SRBs can reach 80% with a flux density above 800 solar flux unit (SFU) at the L2 frequency of GPS. Overall, the detection rate is not affected by the satellite distribution relative to the Sun. The proposed detection method is low-cost and has a high detection rate and low false alarm rate. This method is a noteworthy reference for coping with interference in GNSS from intense L-band SRBs.

DeepAISE – An interpretable and recurrent neural survival model for early prediction of sepsis

Sepsis, a dysregulated immune system response to infection, is among the leading causes of morbidity, mortality, and cost overruns in the Intensive Care Unit (ICU). Early prediction of sepsis can improve situational awareness among clinicians and facilitate timely, protective interventions. While the application of predictive analytics in ICU patients has shown early promising results, much of the work has been encumbered by high false-alarm rates and lack of trust by the end-users due to the ‘black box’ nature of these models. Here, we present DeepAISE (Deep Artificial Intelligence Sepsis Expert), a recurrent neural survival model for the early prediction of sepsis. DeepAISE automatically learns predictive features related to higher-order interactions and temporal patterns among clinical risk factors that maximize the data likelihood of observed time to septic events. A comparative study of four baseline models on data from hospitalized patients at three different healthcare systems indicates that DeepAISE produces the most accurate predictions (AUCs between 0.87 and 0.90) at the lowest false alarm rates (FARs between 0.20 and 0.25) while simultaneously producing interpretable representations of the clinical time series and risk factors.

Forward-Looking Ground-Penetrating Radar: Subsurface target imaging and detection: A review

Detection of shallow-buried, in-road threats using a forward-looking (FL) ground-penetrating radar (GPR) system has attracted significant research interest in the last decade. An FL-GPR mounted on a moving platform can provide standoff target detection and imaging. This enables real-time sensing and situation awareness over large ground areas. The main challenge facing this sensing technology is high false-alarm rates due to scattering arising from air–ground interface roughness and subsurface clutter.

XGB+FM for Severe Convection Forecast and Factor Selection

In the field of meteorology, radiosonde data and observation data are critical for analyzing regional meteorological characteristics. Because of the high false alarm rate, severe convection forecasting is still challenging. In addition, the existing methods are difficult to use to capture the interaction of meteorological factors at the same time. In this research, a cascade of extreme gradient boosting (XGBoost) for feature transformation and a factorization machine (FM) for second-order feature interaction to capture the nonlinear interaction—XGB+FM—is proposed. An attention-based bidirectional long short-term memory (Att-Bi-LSTM) network is proposed to impute the missing data of meteorological observation stations. The problem of class imbalance is resolved by the support vector machines–synthetic minority oversampling technique (SVM-SMOTE), in which two oversampling strategies based on the support vector discrimination mechanism are proposed. It is proven that the method is effective, and the threat score (TS) is 7.27~14.28% higher than other methods. Moreover, we propose the meteorological factor selection method based on XGB+FM and improve the forecast accuracy, which is one of our contributions, as well as the forecast system.

Multisensor-Weighted Fusion Algorithm Based on Improved AHP for Aircraft Fire Detection

Aiming at the high false alarm rate when using single sensor to detect fire in aircraft cabin, a multisensor data fusion method is proposed to detect fire. First, the weights of multiple factors, that is, temperature, smoke concentration, CO concentration, and infrared ray intensity in the event of fire, were calculated by using the improved analytic hierarchy process (AHP) method on each sensor node of wireless sensor network, and the probability of fire event in the cabin was evaluated by multivariable-weighted fusion method. Second, based on the mutual support among the evaluation data of fire probabilities of each node, the adaptive weight coefficient is assigned to each evaluation value, and the weighted fusion of all evaluation values of each node is conducted to obtain the fire probability. In the end, compared to the threshold of probability, the fire alarm is determined. Comparing the proposed algorithm to the grey fuzzy neural network fusion algorithm and fuzzy logic fusion algorithm in terms of the time consumption for fire detection and sending alarm and the accuracy of fire alarm perspectives, the experiments demonstrate that the proposed fire detection algorithm can detect the fire within 10s and reduce the false alarm rate to less than 0.5%, which verifies the superiority of the algorithm in promptness and accuracy. In the meanwhile, the fault tolerance of the algorithm is proved as well.

Ship domain model for ships with restricted manoeuvrability in busy waters

AbstractShip domain is an important theory in ship collision avoidance and an effective collision detection method. First, several classical ship domain models are used in experiments. The results show that the alarm rate is too high in busy waters, leading to greatly reduced practicality of the model. Potential collision risk cannot be detected effectively, especially for a ship with restricted manoeuvrability, which is usually regarded as an overtaken ship due to its navigation characteristics. Therefore, it is necessary to fully consider the interference of other ships to ships with limited manoeuvrability in an encounter situation. A novel ship domain model for ships with restricted manoeuvrability in busy waters is proposed. Considering the navigation characteristics of a ship with restricted manoeuvrability and the influence of the ship–ship effect, an algorithm to determine the boundary of the ship domain model is given by force and moment equations. AIS trajectory data of the North Channel of the Yangtze River Estuary are used to perform a comparative experiment, and four classical ship domain models are employed to perform comparative experiments. The results show that the alarm rates of the novel ship domain model are 7⋅608%, 15⋅131%, 55⋅785% and 7⋅608% lower than those of the other four classical models, and this outcome can effectively reduce the high false alarm rate produced by other models in this environment.

Traffic accident detection and condition analysis based on social networking data

Accurate detection of traffic accidents as well as condition analysis are essential to effectively restoring traffic flow and reducing serious injuries and fatalities. This goal can be obtained using an advanced data classification model with a rich source of traffic information. Several systems based on sensors and social networking platforms have been presented recently to detect traffic events and monitor traffic conditions. However, sensor-based systems provide limited information, and may fail owing to the long detection times and high false-alarm rates. In addition, social networking data are unstructured, unpredictable, and contain idioms, jargon, and dynamic topics. The machine learning algorithms utilized for traffic event detection might not extract valuable information from social networking data. In this paper, a social network–based, real-time monitoring framework is proposed for traffic accident detection and condition analysis using ontology and latent Dirichlet allocation (OLDA) and bidirectional long short-term memory (Bi-LSTM). First, the query-based search engine effectively collects traffic information from social networks, and the data preprocessing module transforms it into structured form. Second, the proposed OLDA-based topic modeling method automatically labels each sentence (e.g., traffic or non-traffic) to identify the exact traffic information. In addition, the ontology-based event recognition approach detects traffic events from traffic-related data. Next, the sentiment analysis technique identifies the polarity of traffic events employing user’s opinions, which helps determine accurate conditions of traffic events. Finally, the FastText model and Bi-LSTM with softmax regression are trained for traffic event detection and condition analysis. The proposed framework is evaluated using traffic-related data, comparing OLDA and Bi-LSTM with existing topic modeling methods and traditional classifiers using word embedding models, respectively. Our system outperforms state-of-the-art methods and achieves accuracy of 97 %. This finding demonstrates that the proposed system is more efficient for traffic event detection and condition analysis, in comparison to other existing systems.

Pipeline leakage monitoring experiments based on evaporation‐enhanced FBG temperature sensing technology

The water distribution systems in utility tunnel are considered as the lifeline projects. The leakage monitoring of water supply and drainage pipelines is closely related to the overall operation and safety of the utility tunnel. However, conventional pipeline leakage monitoring methods are difficult to realize the monitoring of ordinary temperature liquid and small leakage. Moreover, they also have problems with noise interference, high false alarm rate, and high cost. In order to overcome the shortcomings of the conventional methods, the evaporation-enhanced Fiber Bragg Grating (FBG) temperature sensing (EETS-FBG) method was proposed as a new type of leakage monitoring solution according to the principle of the psychrometer. This method utilizes the evaporative cooling characteristics of water-absorbing porous materials and the FBG temperature sensors to realize leakage monitoring. On the basis of experiments, the feasibility of this method is explored, and the effects of wind speed, liquid temperature, and leakage volume on monitoring results are analyzed. The experimental results show that the EETS-FBG method can monitor the temperature difference of 0.97–8.55°C in the humidity of 51.0–90.8%RH, and the measured values are in good agreement with the theoretical values, which shows that the method has good feasibility. In addition, the measured temperature difference increases with the ambient wind speed, while the leakage volume and water temperature have no obvious effect on the temperature difference, which shows that the method has good reliability and universal applicability. This method has outstanding advantages for normal temperature liquid and small leakage monitoring and is worthy of application and promotion.

On Improving Hotspot Detection Through Synthetic Pattern-Based Database Enhancement

Design hotspots are layout patterns which may cause defects due to complex design and process interactions. Several machine learning and pattern matching-based methods have been proposed to identify and correct them early during design stages. However, almost all of them suffer from high false-alarm rates, mainly because they are oblivious to the root causes of hotspots. In this work, we seek to address this limitation by using a novel database enhancement approach through synthetic pattern generation based on a carefully crafted design of experiments. We evaluate the effectiveness of the proposed method using industry-standard tools and designs and demonstrate more than <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$3\times $ </tex-math></inline-formula> reduction in classification error in comparison to the state-of-the-art.

Structural Prior Driven Regularized Deep Learning for Sonar Image Classification

Deep learning has been recently shown to improve performance in the domain of synthetic aperture sonar (SAS) image classification. Given the constant resolution with a range of SAS, it is no surprise that deep learning techniques perform so well. Despite deep learning’s recent success, there are still compelling open challenges in reducing the high false alarm rate and enabling success when training imagery is limited, which is a practical challenge that distinguishes the SAS classification problem from standard image classification set-ups where training imagery may be abundant. We address these challenges by exploiting prior knowledge that humans use to grasp the scene. These include unconscious elimination of the image speckle and localization of objects in the scene. We introduce a new deep learning architecture that incorporates these priors with the goal of improving automatic target recognition (ATR) from SAS imagery. Our proposal—called SPDRDL, structural prior driven regularized deep learning—incorporates the previously mentioned priors in a multitask convolutional neural network (CNN) and requires no additional training data when compared to traditional SAS ATR methods. Two structural priors are enforced via regularization terms in the learning of the network: 1) structural similarity prior—enhanced imagery (often through despeckling) aids human interpretation and is semantically similar to the original imagery and 2) structural scene context priors—learned features ideally encapsulate target centering information; hence learning may be enhanced via a regularization that encourages fidelity against known ground truth target shifts (relative target position from scene center). Experiments on a challenging real-world data set reveal that SPDRDL outperforms state-of-the-art deep learning and other competing methods for SAS image classification.

YOLO-FIRI: Improved YOLOv5 for Infrared Image Object Detection

To solve object detection issues in infrared images, such as a low recognition rate and a high false alarm rate caused by long distances, weak energy, and low resolution, we propose a region-free object detector named YOLO-FIR for infrared (IR) images with YOLOv5 core by compressing channels, optimizing parameters, etc. An improved infrared image object detection network, YOLO-FIRI, is further developed. Specifically, while designing the feature extraction network, the cross-stage-partial-connections (CSP) module in the shallow layer is expanded and iterated to maximize the use of shallow features. In addition, an improved attention module is introduced in residual blocks to focus on objects and suppress background. Moreover, multiscale detection is added to improve small object detection accuracy. Experimental results on the KAIST and FLIR datasets show that YOLO-FIRI demonstrates a qualitative improvement compared with the state-of-the-art detectors. Compared with YOLOv4, the mean average precision (mAP50) of YOLO-FIRI is increased by 21% on the KAIST dataset, the speed is reduced by 62%, the parameters are decreased by 89%, the weight size is reduced by more than 94%, and the computational costs are reduced by 84%. Compared with YOLO-FIR, YOLO-FIRI has an approximately 5% to 20% improvement in AP, AR (average recall), mAP50, F1, and mAP50:75. Furthermore, due to the shortcomings of high noise and weak features, image fusion can be applied to image preprocessing as a data enhancement method by fusing visible and infrared images based on a convolutional neural network.

An LSTM-BINN Approach for Built-In Test Analog Signal State Recognition of Heavy-Duty Gas Turbine Controllers

Built-in test (BIT) is widely employed in equipment state detection. However, the high false alarm rate (FAR) of conventional BIT leads to troubles for manufacturers and users. To solve this problem, a novel BIT analog signal recognition algorithm that integrates long short-term memory (LSTM) and biologically inspired neural network (BINN) is proposed in this article. The effects of signal noise and intermittent faults on BIT results are effectively reduced. The proposed algorithm is mainly used for the BIT analog signal of heavy-duty gas turbine controllers, LSTM is used for preliminary modeling of analog signals, and the BINN is introduced into the extra signal state recognition, serving as a classifier based on stimulation transmission among neurons. Furthermore, the gravitational search algorithm (GSA) is employed to optimize the parameters of the proposed algorithm to improve its performance. By employing BIT temperature signal data from heavy-duty gas turbine controllers, extensive experimental results were given to show that the proposed approach was reasonable and accurate by comparison with conventional BIT. By comparing with other common neural network algorithms, the developed approach can provide lower FAR and more accurate recognition results.

Dim and Small Target Detection in Multi-Frame Sequence Using Bi-Conv-LSTM and 3D-Conv Structure

Infrared dim and small target detection is widely used in military and civil fields. Traditional methods in that application rely on the local contrast between the target and background for single-frame detection. On the other hand, those algorithms depend on the motion model with fixed parameters for multi-frame association. For the great similarity of gray value and the dynamic changes of motion model parameters in the condition of low SNR and strong clutter, those methods possess weak robustness, low detection probability, and high false alarm rate. In this paper, an infrared video sequences encoding and decoding model based on Bidirectional Convolutional Long Short-Term Memory structure (Bi-Conv-LSTM) and 3D Convolutional structure (3D-Conv) is proposed, addressing the problem of high similarity and dynamic changes of parameters. For solving the problem of dynamic change in parameters, Bi-Conv-LSTM structure is used to learn the motion model of targets. And for the problem of low local contrast, 3D-Conv structure is adopted to extend receptive field in the time dimension. In order to improve the precision of detection, the Decoding part is divided into two different full connections with distinctive active function. Simulation results show that the trajectory detection accuracy of the proposed model is more than 90% under the condition of low SNR and maneuvering motion, which is better than traditional method of 80% in DB-TBD 20% in others. Real data experiment to illustrate that that our proposed method can detect small infrared targets of a low false alarm rate and high detection probability.

Multidirectional Ring Top-Hat Transformation for Infrared Small Target Detection

Fast and robust detection of a small target in an infrared image is one of the key techniques in infrared search and track systems. The top-hat transformation is an important technology in the field of infrared target detection. Many algorithms have been proposed to improve the top-hat transformation by modifying different structural elements. However, these methods can cause a high false alarm rate under conditions of small dim targets, low signal-to-noise, and complex background. To overcome this limitation, considering both detection performance and speed, an effective and robust detection model using multidirectional structural elements is proposed in this article. First, the nonconcentric multidirectional ring structural elements are designed. Then, using the morphological transformation with designed structural elements, the background is suppressed, and a small target is enhanced in different directions. Finally, the results of the morphological transformations are fused to obtain the final result, further highlighting a small target. The experimental results demonstrate the robust performance of the proposed method for various backgrounds and targets. In addition, the proposed method can achieve a detection speed of up to 10.4 ms per frame under the image resolution of 240 $ \times $ 320 pixels. Therefore, the proposed method is suitable for real-time applications.

Cost-Sensitive Siamese Network for PCB Defect Classification.

After the production of printed circuit boards (PCB), PCB manufacturers need to remove defected boards by conducting rigorous testing, while manual inspection is time-consuming and laborious. Many PCB factories employ automatic optical inspection (AOI), but this pixel-based comparison method has a high false alarm rate, thus requiring intensive human inspection to determine whether alarms raised from it resemble true or pseudo defects. In this paper, we propose a new cost-sensitive deep learning model: cost-sensitive siamese network (CSS-Net) based on siamese network, transfer learning and threshold moving methods to distinguish between true and pseudo PCB defects as a cost-sensitive classification problem. We use optimization algorithms such as NSGA-II to determine the optimal cost-sensitive threshold. Results show that our model improves true defects prediction accuracy to 97.60%, and it maintains relatively high pseudo defect prediction accuracy, 61.24% in real-production scenario. Furthermore, our model also outperforms its state-of-the-art competitor models in other comprehensive cost-sensitive metrics, with an average of 33.32% shorter training time.

Change Point Detection in Time Series Data Using Autoencoders With a Time-Invariant Representation

Change point detection (CPD) aims to locate abrupt property changes in time series data. Recent CPD methods demonstrated the potential of using deep learning techniques, but often lack the ability to identify more subtle changes in the autocorrelation statistics of the signal and suffer from a high false alarm rate. To address these issues, we employ an autoencoder-based methodology with a novel loss function, through which the used autoencoders learn a partially time-invariant representation that is tailored for CPD. The result is a flexible method that allows the user to indicate whether change points should be sought in the time domain, frequency domain or both. Detectable change points include abrupt changes in the slope, mean, variance, autocorrelation function and frequency spectrum. We demonstrate that our proposed method is consistently highly competitive or superior to baseline methods on diverse simulated and real-life benchmark data sets. Finally, we mitigate the issue of false detection alarms through the use of a postprocessing procedure that combines a matched filter and a newly proposed change point score. We show that this combination drastically improves the performance of our method as well as all baseline methods.

A Spectrogram Image-Based Network Anomaly Detection System Using Deep Convolutional Neural Network

The dynamics of computer networks have changed rapidly over the past few years due to a tremendous increase in the volume of the connected devices and the corresponding applications. This growth in the network's size and our dependence on it for all aspects of our life have therefore resulted in the generation of many attacks on the network by malicious parties that are either novel or the mutations of the older attacks. These attacks pose many challenges for network security personnel to protect the computer and network nodes and corresponding data from possible intrusions. A network intrusion detection system (NIDS) can act as one of the efficient security solutions by constantly monitoring the network traffic to secure the entry points of a network. Despite enormous efforts by researchers, NIDS still suffers from a high false alarm rate (FAR) in detecting novel attacks. In this paper, we propose a novel NIDS framework based on a deep convolution neural network that utilizes network spectrogram images generated using the short-time Fourier transform. To test the efficiency of our proposed solution, we evaluated it using the CIC-IDS2017 dataset. The experimental results have shown about 2.5% - 4% improvement in accurately detecting intrusions compared to other deep learning (DL) algorithms while at the same time reducing the FAR by 4.3%-6.7% considering binary classification scenario. We also observed its efficiency for a 7-class classification scenario by achieving almost 98.75% accuracy with 0.56% - 3.72% improvement compared to other DL methodologies.