In general, a constant false alarm rate algorithm (CFAR) is widely used to automatically detect targets in an automotive frequency-modulated continuous wave (FMCW) radar system. However, if the number of guard cells, ...
详细信息
In general, a constant false alarm rate algorithm (CFAR) is widely used to automatically detect targets in an automotive frequency-modulated continuous wave (FMCW) radar system. However, if the number of guard cells, the number of training cells, and the probability of false alarm are set improperly in the conventional CFAR algorithm, the target detection performance is severely degraded. Therefore, we propose a method using a convolutional neural network-based autoencoder (AE) to replace the CFAR algorithm in the multiple-input and multiple-output FMCW radar system. In the AE, the entire detection result is compressed at the encoder side, and only significant signal components are recovered on the decoder side. In this work, by changing the number of hidden layers and the number of filters in each layer, the structure of the AE showing a high signal-to-noise ratio in the target detection result is determined. To evaluate the performance of the proposed method, the AE-based target detection result is compared with the target detection results of conventional CFAR algorithms. As a result of calculating the correlation coefficient with the data marked with the actual target position, the proposed AE-based target detection shows the highest similarity with a correlation of 0.73 or higher.
This study utilizes the pre-processed fMRI data provided by the 2024 ICBHI challenge and excludes two other signal data types with missing information to construct a CNN model that can distinguish three emotion classe...
详细信息
ISBN:
(纸本)9783031863226;9783031863233
This study utilizes the pre-processed fMRI data provided by the 2024 ICBHI challenge and excludes two other signal data types with missing information to construct a CNN model that can distinguish three emotion classes and their corresponding levels. However, due to the high variability and noise in the pre-processed fMRI data, a simple CNN model alone cannot achieve good classification performance. This study addresses the issue of noisy data by proposing a two-stage deep learning model training framework. In the first stage, an autoencoder method is adopted, leveraging its ability to effectively encode and decode data to extract useful signal features from the noisy data for use in the subsequent second stage. In the second stage, the effective features obtained by the encoder are transferred, and the weights of the encoding layers are combined with a fully connected layer for model retraining. This study also analyzes different methods of transferring the weights of the encoding layers. The best model for this study achieved an error rate of only 0.3383 on the official evaluation metric.
Evaluation measures of Generative Adversarial Networks (GANs) have been an active area of research and, currently, there are several measures to evaluate them. The most used GANs evaluation measure is the Frechet Ince...
详细信息
Evaluation measures of Generative Adversarial Networks (GANs) have been an active area of research and, currently, there are several measures to evaluate them. The most used GANs evaluation measure is the Frechet Inception Distance (FID). Measures such as FID are known as model-agnostic methods, where the generator is used as a black box to sample the generated images. Like other measures of model-agnostic, FID uses a deep supervised model for mapping real and generated samples to a feature space. We proposed an approach here with a deep unsupervised model, the Vector Quantised-Variational autoencoder (VQ-VAE), for estimating the mean and the covariance matrix of the Frechet Distance and named it Frechet autoencoder Distance (FAED). Our experimental results highlighted that the feature space of the VQ-VAE describes a clustering domain-specific representation more intuitive and visually plausible than the Inception network used by the benchmark FID.
In many developed countries, photovoltaic solar power, which is considered the most cost-effective renewable energy source, accounts for a major portion of electricity production. The photovoltaic (PV) power generatio...
详细信息
In many developed countries, photovoltaic solar power, which is considered the most cost-effective renewable energy source, accounts for a major portion of electricity production. The photovoltaic (PV) power generation is unpredictable and imprecise due to its high variation that can be caused of meteorological elements, to reduce the negative influence of the use of PV power, accurate PV power prediction is of crucial significance for the secure and efficient operation of photovoltaic power system operation. In light of this, we propose a long short-term memory (LSTM) autoencoder (AE) for photovoltaic power forecasting. Initially, to generate encoded sequences the LSTM-encoder extracts the characteristics from the input data. Then the LSTM-decoder decoded the encoded sequences to advance them to the last dense layer for photovoltaic power prediction. Furthermore, we conducted experiments using a 23.40 kW PV power plants dataset from DKASC in Australia. The results have confirmed that the LSTM-AE model can achieve better prediction accuracy than the benchmark deep learning methods, in terms of mean absolute error (MAE), mean square error (MSE), mean bias error (MBE), root mean square error (RMSE) and coefficient of determination (R-2) performance measures. When the results are analyzed, the LSTM-AE model gives the best results with average RMSE, MAE, and R-2 to 0.0762 kW, 0.0389 kW, and 99.93%, respectively. The experimental results confirm that proposed method with the highest R-2 values and minimum forecasting errors compared to the benchmark models can effectively improve stable performance and achieve better accurate photovoltaic power forecasting.
In this letter, we propose a feedback reduction scheme based on autoencoder and deep reinforcement learning for frequency division duplex (FDD) overlay device-to-device (D2D) communication networks. All D2D receivers ...
详细信息
In this letter, we propose a feedback reduction scheme based on autoencoder and deep reinforcement learning for frequency division duplex (FDD) overlay device-to-device (D2D) communication networks. All D2D receivers and transmitters are equipped with an Encoder and Decoder, respectively, of the trained autoencoder. The D2D receivers compress the feedback information using the Encoder before transmitting while the transmitters decompress the received feedback information. We also employ a dueling deep Q network (DQN) to allow each D2D transmitter to autonomously determine whether to transmit data based on the decompressed feedback information. The performance of the proposed feedback reduction scheme is analyzed in terms of the average MSE of the autoencoder and the average sum-rate of a D2D communication network. Our numerical results show that the proposed feedback reduction scheme using the autoencoder can achieve 100%, 89%, and 86% of the average sum-rate of the perfect feedback scheme with no compression when the signal-to-noise ratio is 10dB, -5dB, and -20dB, respectively, while reducing the feedback by 50%.
Recommender systems are crucial in the big data era, effectively mitigating information overload. Existing recommendation methods are limited on highly sparse data and have mediocre recall performance. Group influence...
详细信息
Recommender systems are crucial in the big data era, effectively mitigating information overload. Existing recommendation methods are limited on highly sparse data and have mediocre recall performance. Group influence aggregates knowledge from different users or organizations to generate decisions, improving information fusion efficiency and group decision-making quality. In this paper, a group influence-based deep adversarial autoencoder (GI-AAE), is proposed for top-N recommendation. It leverages group influence to strengthen autoencoder latent features and address sparse data and uses adversarial learning from GANs to enhance reconstruction. The group influence based deep autoencoder (GI-AE) is the generative model for the GI-AAE. Experimental results show that the proposed algorithm is competitive and gets higher recall values.
Image generation has seen huge leaps in the last few years. Less than 10 years ago we could not generate accurate images using deep learning at all, and now it is almost impossible for the average person to distinguis...
详细信息
ISBN:
(纸本)9798400701870
Image generation has seen huge leaps in the last few years. Less than 10 years ago we could not generate accurate images using deep learning at all, and now it is almost impossible for the average person to distinguish a real image from a generated one. In spite of the fact that image generation has some amazing use cases, it can also be used with ill intent. As an example, deepfakes have become more and more indistinguishable from real pictures and that poses a real threat to society. It is important for us to be vigilant and active against deepfakes, to ensure that the false information spread is kept under control. In this context, the need for good deepfake detectors feels more and more urgent. There is a constant battle between deepfake generators and deepfake detection algorithms, each one evolving at a rapid pace. But, there is a big problem with deepfake detectors: they can only be trained on so many data points and images generated by specific architectures. Therefore, while we can detect deepfakes on certain datasets with near 100% accuracy, it is sometimes very hard to generalize and catch all real-world instances. Our proposed solution is a way to augment deepfake detection datasets using deep learning architectures, such as autoencoders or U-Net. We show that augmenting deepfake detection datasets using deep learning improves generalization to other datasets. We test our algorithm using multiple architectures, with experimental validation being carried out on state-of-the-art datasets like CelebDF and DFDC Preview. The framework we propose can give flexibility to any model, helping to generalize to unseen datasets and manipulations.
High-resolution range profile (HRRP) is indispensable for modern radar automatic target recognition (RATR) systems and commonly has noisy background and misalignment along the range dimension. In this article, we prop...
详细信息
High-resolution range profile (HRRP) is indispensable for modern radar automatic target recognition (RATR) systems and commonly has noisy background and misalignment along the range dimension. In this article, we propose a novel algorithm for HRPP recognition with the improvement of recognition accuracy, stability against range translation, and robustness to noise background. This method divides the HRRP into patches and encodes the patches to latent feature vectors via a multihead attention mechanism, modeling the spatial dependence, eliminating the influence of range translation, and focusing on the target area. A decoder is adopted to reconstruct the HRRP and improve the robustness against the noise. Experiments on measured data prove that the proposed algorithm outperforms other existing methods with recognition accuracy increased significantly, sensitivity to the range translation eliminated, and noise robustness improved. This study provides a promising and effective approach for HRRP target recognition.
Hyperspectral anomaly detection (HAD) plays a vital role in military and civilian applications. However, compared with target detection or classification tasks, HAD is more challenging due to insufficient anomaly info...
详细信息
Hyperspectral anomaly detection (HAD) plays a vital role in military and civilian applications. However, compared with target detection or classification tasks, HAD is more challenging due to insufficient anomaly information and the difficulty of extracting local and global discriminative features. In this letter, a convolutional transformer-inspired autoencoder (CTA) is proposed for HAD. The CTA consists of a clustering-based module and an autoencoder-based module. First, note that the number of anomalies is small, and distinct from their surroundings, a clustering-based module is proposed to detect the pseudo-background and anomaly samples. Second, the autoencoder module is composed of an encoder and a decoder formed from several skip-connected convolutions and multihead attention-based transformers. The CTA is trained not only to distinguish the anomalies from the background but also to reconstruct the input hyperspectral images (HSIs). Benefiting from integrating the convolution and transformer, the CTA has local and global receptive fields. Moreover, both background and anomaly information explored by the clustering-based module can be adopted to improve the separability of anomalies. Experiments on two hyperspectral datasets demonstrate that the proposed CTA achieves superior detection performance to its counterparts. The code is available at https://***/hzhdhz/CTA.
Graph anomaly detection, aimed at identifying anomalous patterns that significantly differ from other nodes, has drawn widespread attention in recent years. Due to the complex topological structures and attribute info...
详细信息
ISBN:
(纸本)9789819755714;9789819755721
Graph anomaly detection, aimed at identifying anomalous patterns that significantly differ from other nodes, has drawn widespread attention in recent years. Due to the complex topological structures and attribute information inherent in graphs, conventional methods often struggle to effectively identify anomalies. Deep anomaly detection methods based on Graph Neural Networks (GNNs) have achieved significant success. However, they face the challenge of not only obtaining limited neighborhood information but over-smoothing. Over-smoothing is the phenomenon where the representations of nodes gradually become similar and flattened across multiple convolutional layers, thereby limiting the comprehensive learning of neighborhood information. Therefore, we propose a novel anomaly detection framework, TransGAD, to address these challenges. Inspired by the Graph Transformer, we introduce a Transformer-based autoencoder. Treating each node as a sequence and its neighborhood as tokens in the sequence, this autoencoder captures both local and global information. We incorporate cosine positional encoding and masking strategy to obtain more informative node representations and leverage reconstruction error for improved anomaly detection. Experimental results on seven datasets demonstrate that our approach outperforms the state-of-the-art methods.
暂无评论