One of the most prominent security challenges to neural networks are adversarial examples - inputs with often barely perceptible perturbations causing misclassification. In this study, we propose a defense mechanism t...
详细信息
ISBN:
(纸本)9783031789793;9783031789809
One of the most prominent security challenges to neural networks are adversarial examples - inputs with often barely perceptible perturbations causing misclassification. In this study, we propose a defense mechanism that uses an autoencoder to restore adversarial examples before classification. That is, the autoencoder purifies input data points from potential adversarial perturbations. The method is titled autoencoder-based Adversarial Purification (AAP). We demonstrate the effectiveness of AAP on multiple datasets, attack methods, and perturbation levels. While certain limitations exist, this research offers valuable insights and a promising direction for robust defense mechanisms in adversarial deep learning.
In the present scenario, Electrocardiogram (ECG) is an effective non-invasive clinical tool, which reveals the functionality and rhythm of the heart. The non-stationary nature of ECG signal, noise existence, and heart...
详细信息
In the present scenario, Electrocardiogram (ECG) is an effective non-invasive clinical tool, which reveals the functionality and rhythm of the heart. The non-stationary nature of ECG signal, noise existence, and heartbeat abnormality makes it difficult for clinicians to diagnose arrhythmia. The most of the existing models concentrate only on classification accuracy. In this manuscript, an automated model is introduced that concentrates on arrhythmia type classification using ECG signals, and also focuses on computational complexity and time. After collecting the signals from the MIT-BIH database, the signal transformation and decomposition are performed by Multiscale Local Polynomial Transform (MLPT) and Ensemble Empirical Mode Decomposition (EEMD). The decomposed ECG signals are given to the feature extraction phase for extracting features. The feature extraction phase includes six techniques: standard deviation, zero crossing rate, mean curve length, Hjorth parameters, mean Teager energy, and log energy entropy. Next, the feature dimensionality reduction and arrhythmia classification are performed utilizing the improved Firefly Optimization Algorithm and autoencoder. The selection of optimal feature vectors by the improved Firefly Optimization Algorithm reduces the computational complexity to linear and consumes computational time of 18.23 seconds. The improved Firefly Optimization Algorithm and autoencoder model achieved 98.96% of accuracy in the arrhythmia type classification, which is higher than the comparative models.
Image retrieval systems have long been a cornerstone of multimedia databases, aiming to extract relevant images from vast datasets. Batik, an intricate traditional art form originating from Indonesia, presents unique ...
详细信息
ISBN:
(纸本)9798350348798;9798350348804
Image retrieval systems have long been a cornerstone of multimedia databases, aiming to extract relevant images from vast datasets. Batik, an intricate traditional art form originating from Indonesia, presents unique challenges in this domain. Its diverse patterns and rich history have led to a broad array of designs, making the retrieval of specific Batik images a formidable task. This research paper introduces a novel approach that employs a convolutional autoencoder to address the challenges of Batik image retrieval. The underlying problem we address is the difficulty in obtaining high precision results from conventional image retrieval systems when dealing with the detailed patterns of Batik. Our methodology hinges on a convolutional autoencoder, a deep learning model adept at extracting pivotal features from images and using them for various tasks, including image retrieval. The effectiveness of our method was evaluated using the Batik Nitik dataset, a comprehensive collection of 960 images representing a wide range of Batik designs. The outcomes were promising. Our proposed method achieved a remarkable precision of 0.98, a testament to its capability in accurately retrieving relevant Batik images. Furthermore, this research not only demonstrates the potential of convolutional autoencoders in the realm of image retrieval but also offers a solution tailored to the unique intricacies of Batik. In conclusion, the convolutional autoencoder presents a groundbreaking approach to Batik image retrieval, merging traditional art with modern deep learning techniques to ensure accuracy and relevance in results.
Large occlusions result in a significant decline in image classification accuracy. During inference, diverse types of unseen occlusions introduce out-of-distribution data to the classification model, leading to accura...
详细信息
ISBN:
(纸本)9798350349405;9798350349399
Large occlusions result in a significant decline in image classification accuracy. During inference, diverse types of unseen occlusions introduce out-of-distribution data to the classification model, leading to accuracy dropping as low as 50%. As occlusions encompass spatially connected regions, conventional methods involving feature reconstruction are inadequate for enhancing classification performance. We introduce LEARN: Latent Enhancing feAture Reconstruction Network- An auto-encoder based network that can be incorporated into the classification model before its classifier head without modifying the weights of classification model. In addition to reconstruction and classification losses, training of LEARN effectively combines intra- and inter-class losses calculated over its latent space-which lead to improvement in recovering latent space of occluded data, while preserving its class-specific discriminative information. On the OccludedPASCAL3D+ dataset, the proposed LEARN outperforms standard classification models (VGG16 and ResNet-50) by a large margin and up to 2% over state-of-the-art methods. In cross-dataset testing, our method improves the average classification accuracy by more than 5% over the state-of-the-art methods. In every experiment, our model consistently maintains excellent accuracy on in-distribution data.
A variety of ranging threats represented by Ghost Peak attack have raised concerns regarding the security performance of Ultra-Wide Band (UWB) systems with the finalization of the IEEE 802.15.4z standard. Based on cha...
详细信息
ISBN:
(纸本)9798350378412
A variety of ranging threats represented by Ghost Peak attack have raised concerns regarding the security performance of Ultra-Wide Band (UWB) systems with the finalization of the IEEE 802.15.4z standard. Based on channel reciprocity, this paper proposes a low complexity attack detection scheme that compares Channel Impulse Response (CIR) features of both ranging sides utilizing an autoencoder with the capability of data compression and feature extraction. Taking Ghost Peak attack as an example, this paper demonstrates the effectiveness, feasibility and generalizability of the proposed attack detection scheme through simulation and experimental validation. The proposed scheme achieves an attack detection success rate of over 99% and can be implemented in current systems at low cost.
Photovoltaic inverter health prediction is a crucial aspect of the reliability and performance maintenance of photovoltaic power generation systems. With the rapid development of solar energy, the inverter, as one of ...
详细信息
ISBN:
(纸本)9798400716638
Photovoltaic inverter health prediction is a crucial aspect of the reliability and performance maintenance of photovoltaic power generation systems. With the rapid development of solar energy, the inverter, as one of the core components of photovoltaic power generation systems, plays a vital role in ensuring the effective conversion of energy. Traditional methods for predicting the health of photovoltaic inverters involve simple weighted summation of device-generated data or basic classification assessments. These approaches often lack precision in predicting device health. This paper proposes a data-driven health prediction method that integrates operational environment data from photovoltaic inverters with performance data during operation. Different autoencoders are trained as environmental benchmark models based on various working conditions. Real-time operational data is input into the health model to generate health scores reflecting the device's condition. Experimental results demonstrate that the constructed health model effectively fits the dataset and accurately assesses the operating status of photovoltaic inverters. By enabling real-time health assessment and prompt maintenance actions, this method provides an effective guarantee for increasing photovoltaic power generation efficiency, potentially significantly reducing maintenance costs, and enhancing system reliability and maintainability. This, in turn, contributes significantly to the sustainable development of renewable energy in the field.
Women experience major bodily changes both during pregnancy and post-pregnancy. Diastasis Recti Abdominis (DRA) is a noticeable issue in the postpartum period among the female population in the world. Though postnatal...
详细信息
Women experience major bodily changes both during pregnancy and post-pregnancy. Diastasis Recti Abdominis (DRA) is a noticeable issue in the postpartum period among the female population in the world. Though postnatal fitness has gained attention in the recent decade, there is scarce knowledge of the abnormal condition called DRA and its consequences. In the presence of an abnormality, women feel less energetic in their daily activities and may experience fatigue in the abdominal muscles. The physical way of regaining strength in core abdominal muscles includes rehabilitation through exercises prescribed by physiotherapists. The sit-up and curl-up exercises engage the core abdominal muscles and when practiced regularly can bring back the separated recti muscles together in time. In order to bring this practice unsupervised by the physicians and monitor the pace of exercises by the patient individually, wearable Inertial measurement unit (IMU) sensors were employed. The utilization of IMU wearable sensors for DRA has been sparsely explored in literature. In this study, two groups of subjects with DRA perform the rehabilitation exercises and respective inertial measurements were observed. When the situation goes unsupervised, the effective contraction of the abdominal recti muscles and the correctness of exercises were uncertain. It's a well-known fact that deep learning algorithms aid in determining the significant features thereby making the unsupervised classification problem more efficient. Here in this study an ensembled autoencoder neural network is implemented in which the IMU datasets were employed for the classification of correct and incorrect exercises. The latent vector generated in the autoencoder model encapsulates the inherent patterns of the input by undertaking all occurrences into a latent space. Thereby in this work, the reconstruction error generated from the autoencoder network is used to determine the correct and incorrect exercise. The ensemble
Linear local tangent space alignment (LLTSA) is a classical dimensionality reduction method based on manifold. However, LLTSA and all its variants only consider the one-way mapping from high-dimensional space to low-d...
详细信息
Linear local tangent space alignment (LLTSA) is a classical dimensionality reduction method based on manifold. However, LLTSA and all its variants only consider the one-way mapping from high-dimensional space to low-dimensional space. The projected low-dimensional data may not accurately and effectively "represent" the original samples. This paper proposes a novel LLTSA method based on the linear autoencoder called LLTSA-AE (LLTSA with autoencoder). The proposed LLTSA-AE is divided into two stages. The conventional process of LLTSA is viewed as the encoding stage, and the additional and important decoding stage is used to reconstruct the original data. Thus, LLTSA-AE makes the low-dimensional embedding data "represent" the original data more accurately and effectively. LLTSA-AE gets the recognition rates of 85.10, 67.45, 75.40 and 86.67% on handwritten Alphadigits, FERET, Georgia Tech. and Yale datasets, which are 9.4, 14.03, 7.35 and 12.39% higher than that of the original LLTSA respectively. Compared with some improved methods of LLTSA, it also obtains better performance. For example, on Handwritten Alphadigits dataset, compared with ALLTSA, OLLTSA, PLLTSA and WLLTSA, the recognition rates of LLTSA-AE are improved by 4.77, 3.96, 7.8 and 8.6% respectively. It shows that LLTSA-AE is an effective dimensionality reduction method.
The high-flux advanced neutron application reactor (HANARO) is a multi-purpose research reactor at the Korea Atomic Energy Research Institute (KAERI). HANARO has been used in scientific and industrial research and dev...
详细信息
The high-flux advanced neutron application reactor (HANARO) is a multi-purpose research reactor at the Korea Atomic Energy Research Institute (KAERI). HANARO has been used in scientific and industrial research and developments. Therefore, stable operation is necessary for national science and industrial prospects. This study proposed an anomaly detection system based on deep learning, that supports the stable operation of HANARO. The proposed system collects multiple sensor data, displays system information, analyzes status, and performs anomaly detection using deep autoencoder. The system comprises communication, visualization, and anomaly-detection modules, and the prototype system is implemented on site in 2021. Finally, an analysis of the historical data and synthetic anomalies was conducted to verify the overall system;simulation results based on the historical data show that 12 cases out of 19 abnormal events can be detected in advance or on time by the deep learning AD model.(c) 2023 Korean Nuclear Society, Published by Elsevier Korea LLC. This is an open access article under the CC BY-NC-ND license (http://***/licenses/by-nc-nd/4.0/).
The use of Low-Power Wide Area Network (LPWAN) technologies, such as Long Range (LoRa), Sigfox, and IEEE 802.15.4g (ZigBee), has grown significantly, addressing a wide range of applications including smart metering, a...
详细信息
The use of Low-Power Wide Area Network (LPWAN) technologies, such as Long Range (LoRa), Sigfox, and IEEE 802.15.4g (ZigBee), has grown significantly, addressing a wide range of applications including smart metering, agriculture, smart homes, and healthcare. These technologies are valued for their simplicity, flexible connectivity, low power consumption, efficient modulation techniques, and moderate data rates. As a result, they can coexist within the same environment, serving either similar or distinct applications. However, the increasing deployment of devices and technologies has amplified the likelihood of interference between them, leading to performance degradation, particularly in real-world scenarios under challenging conditions where noise power surpasses signal power. The rapid proliferation of these technologies, especially within unlicensed Industrial, Scientific, and Medical (ISM) frequency bands, underscores the need for effective techniques to ensure seamless coexistence without disrupting communication. To address this challenge, we investigate the role of data representation and propose a Channel Attention-based Denoising autoencoder U-Net and Classifier (UNA-DAEC). This model is designed to denoise multi-label LPWAN signals affected by white Gaussian noise and accurately classify overlapping transmissions, specifically IEEE 802.15.4g, Sigfox, and LoRa signals, within the same environment. The primary objective of UNA-DAEC is to achieve reliable signal classification in low Signal-to-Noise Ratio (SNR) conditions. This is achieved by first denoising the noisy signals to obtain optimal representations, enabling high classification accuracy with a single forward and backward propagation. Our results further demonstrate that data representation plays a critical role in identifying and classifying LPWAN signals, particularly in challenging low-SNR environments, with a significant performance of 44%, 7%, and 26% over CNN-based IQ, CNN-based FFT and DAE+Cla
暂无评论