Frequency Domain Transformation Research Articles

SpanEffiDet: Span-Scale and Span-Path Feature Fusion for Object Detection

Lower versions of EfficientDet (such as D0, D1) have smaller network structures and parameter sizes, but lower detection accuracy. Higher versions exhibit higher accuracy, but the increase in network complexity poses challenges for real-time processing and hardware requirements. To meet the higher accuracy requirements under limited computational resources, this paper introduces SpanEffiDet based on the channel adaptive frequency filter (CAFF) and the Span-Path Bidirectional Feature Pyramid structure. Firstly, the CAFF module proposed in this paper realizes the frequency domain transformation of channel information through Fourier transform and effectively extracts the key features through semantic adaptive frequency filtering, thus, eliminating channel redundant information of EfficientNet. Simultaneously, the module has the ability to compute the weights across the channels and at fine granularity, and capture the detailed information of element features. Secondly, a two-way characteristic pyramid network multi-level cross-BIFPN, which can achieve multi-layer and multi-nodes, is proposed to build cross-level information transmission to incorporate both semantic and positional information of the target. This design enables the network to more effectively detect objects with significant size differences in complex environments. Finally, by introducing generalized focal Loss V2, reliable localization quality estimation scores are predicted from the distribution statistics of bounding boxes, thereby improving localization accuracy. The experimental results indicate that on the MS COCO dataset, SpanEffiDet-D0 achieved an AP improvement of 3.3% compared to the original EfficientDet series algorithms. Similarly, on the PASCAL VOC2007 and 2012 datasets, the mAP of SpanEffiDet-D0 is respectively 1.66 and 2.65% higher than that of EfficientDet-D0.

Denoising and analysis of synthetic aperture radar images using improved weight threshold technique in curvelet transform frequency domain

Denoising and analysis of synthetic aperture radar images using improved weight threshold technique in curvelet transform frequency domain

Efficient Cyclostationary Spectrum Sensing Using Low Complexity FFT Algorithms

One of the major problems of cyclostationary spectrum sensing (CSS) system in cognitive radios (CR) are the implementation complexity. One possible way to reduce CSS complexity is to use efficient algorithms for performing Fast Fourier Transformation (FFT). Over the years, a lot of different FFT algorithms have been created. This includes the Split-Radix algorithm., the Fast Hartley Transform (FHT), and slide DFT. This paper investigates the suitable FFT algorithm among the aforementioned techniques, cyclostationary feature detection (CFD)-based spectrum sensing stands out. The methods have been thoroughly compared based on computational time, object size, code size, data dependence (real or complex), and the amount of mathematical operations involved in the computations. Simulation results show that slide FFT is the suitable frequency domain transformation algorithm to use in implementing cyclostationary spectrum sensing in cognitive radios as compared to the other considered algorithms where it provides a significant reduction in FFT stage computation complexity reach to 17% in SRFFT , 78% in FHT and 82% in SDFT while keeping the detection probability at satisfactory level.

Frequency-domain image encryption based on IWT and 3D S-box

Most of the existing spatial domain image encryption techniques suffer from the difficulty of resisting cryptographic attacks. For this reason, this paper proposes a frequency-domain based digital image encryption scheme by combining Integer Wavelet Transform (IWT), three-dimensional S-box and chaotic system. First, the plaintext image is decomposed into different frequency subbands by IWT to map the digital image from spatial domain to frequency domain. Second, the plaintext hash value is selected as the dynamic key, and dynamic chaotic pseudo-random sequences with associations are generated, which are used for the encryption of each module respectively. Then, a three-dimensional S-box is designed to encrypt the information-rich low-frequency information using ‘bit-permutation three-dimensional S-box replace ciphertext interleaved diffusion’, while the high-frequency information is encrypted using a lightweight ‘XOR-row column permutation’ operation. Finally, the secure ciphertext for public channel transmission is obtained by the reconstruction method. The scheme of this paper, the frequency domain transformation is implemented through IWT, which enhances the ability to resist attacks. In addition, the diffusion encryption modules employ the introduction of ciphertext interleaved diffusion and parallel encryption mechanisms, thus the algorithm has the ability to resist plaintext attacks. Theoretical analysis and empirical results show that the algorithm has excellent numerical statistical analysis results, which corroborate that it has good confusion, diffusion and avalanche effects, and is able to resist various common cryptographic attacks. The frequency domain image encryption scheme proposed in this paper is a preferred high-security digital image privacy protection technique, so it has good application prospects.

Frequency domain model-augmented feature space transferable attack

Deep neural networks (DNNs) are susceptible to adversarial attacks, and one important factor is that adversarial samples are transferable, i.e., adversarial samples generated by a particular network may deceive other black-box models. However, existing transferable adversarial attacks tend to modify the input features of images directly without selection to reduce the prediction accuracy in the alternative model, which would enable the adversarial samples to fall into the model’s local optimum. Alternative models differ significantly from the victim model in most cases, and while simultaneously attacking multiple models may improve transferability, gathering numerous different models is more challenging and expensive. We simulate various models using frequency domain transformation to close the gap between the source and victim models and improve transferability. At the same time, we destroy important intermediate layer features that influence the decision of the model in the feature space. Additionally, smoothing loss is introduced to remove high-frequency perturbations. Extensive experiments demonstrate that our FM-FSTA attack generates more well-hidden and transferable adversarial samples, and achieves a high deception rate even when attacking adversarially trained models. Compared to other methods, our FM-FSTA improved attack success rate under different defense mechanisms, which reveals the potential threats of current robust models.

MMDCP: An Image Enhancement Algorithm Incorporating Multi-Channel Phase Activation and Multi-Constrained Dark Channel Prior

The quality of visual media is critically impacted by low illumination and the presence of airborne particulates, leading to challenges in brightness balance, color saturation, and texture clarity which are detrimental to various applications in image processing and computer vision. Addressing these challenges, this study introduces a novel image enhancement algorithm that significantly improves the quality of degraded images. Our proposed method, the multi-channel phase activation and multi-constraint dark channel prior (MMDCP), leverages an innovative approach by integrating the phase-adjusted Gaussian kernel function for brightness channel optimization in the Fourier transform frequency domain. This optimization is enhanced through the application of a saturated dark channel prior, achieving simultaneous brightness enhancement and color fidelity. Furthermore, we refine the dark channel prior deblurring algorithm by incorporating intensity, brightness, and color constraints to correct overexposure issues and color offsets in the reconstructed images. The efficacy of the MMDCP algorithm is demonstrated through extensive experimentation, comparing it against six contemporary image enhancement algorithms using two types of objective indicators and subjective assessments across four public datasets. The MMDCP algorithm consistently outperforms the existing methods, with a notable average improvement of 20% in PSNR and 19.6% in SSIM metrics, substantiating its superiority in enhancing brightness, detail, and color accuracy. This study’s results underline the MMDCP algorithm’s robustness and versatility in improving image quality across various conditions, including daytime, nighttime, indoor, and outdoor settings.

Improving Equity in Deep Learning Medical Applications with the Gerchberg-Saxton Algorithm

AbstractDeep learning (DL) has gained prominence in healthcare for its ability to facilitate early diagnosis, treatment identification with associated prognosis, and varying patient outcome predictions. However, because of highly variable medical practices and unsystematic data collection approaches, DL can unfortunately exacerbate biases and distort estimates. For example, the presence of sampling bias poses a significant challenge to the efficacy and generalizability of any statistical model. Even with DL approaches, selection bias can lead to inconsistent, suboptimal, or inaccurate model results, especially for underrepresented populations. Therefore, without addressing bias, wider implementation of DL approaches can potentially cause unintended harm. In this paper, we studied a novel method for bias reduction that leverages the frequency domain transformation via the Gerchberg-Saxton and corresponding impact on the outcome from a racio-ethnic bias perspective.

A Multi Representation Deep Learning Approach for Epileptic Seizure Detection

Epileptic seizures, unpredictable in nature and potentially dangerous during activities like driving, pose significant risks to individual and public safety. Traditional diagnostic methods, which involve labour-intensive manual feature extraction from Electroencephalography (EEG) data, are being supplanted by automated deep learning frameworks. This paper introduces an automated epileptic seizure detection framework utilizing deep learning to bypass manual feature extraction. Our framework incorporates detailed pre-processing techniques: normalization via L2 normalization, filtering with an 80 Hz and 0,5 Hz Butterworth low-pass and high-pass filter, and a 50 Hz IIR Notch filter, channel selection based on standard deviation calculations and Mutual Information algorithm, and frequency domain transformation using FFT or STFT with Hann windows and 50% hop. We evaluated on two datasets: the first comprising 4 canines and 8 patients with 2.299 ictal, 23.445 interictal, and 32.915 test data, 400-5000Hz sampling rate across 16-72 channels; the second dataset, intended for testing, 733 icatal, 4.314 interictal, and 1908 test data, each 10 minutes long, recorded at 400Hz across 16 channels. Three deep learning architectures were assessed: CNN, LSTM, and a hybrid CNN-LSTM model-stems from their demonstrated efficacy in handling the complex nature of EEG data. Each model offers unique strengths, with the CNN excelling in spatial feature extraction, LSTM in temporal dynamics, and the hybrid model combining these advantages. The CNN model, comprising 31 layers, yielded highest accuracy, achieving 91% on the first dataset (precision 92%, recall 91%, F1-score 91%) and 82% on the second dataset using a 30-second threshold. This threshold was chosen for its clinical relevance. The research advances epileptic seizure detection using deep learning, indicating a promising direction for future medical technology. Future work will focus on expanding dataset diversity and refining methodologies to build upon these foundational results.

Enhancing response estimation and system identification in structural health monitoring through data-driven approaches

Through the advancement of Data Science methodologies, a new era in output-only identification techniques has been inaugurated, driven by the integration of data-driven methodologies within the realm of Structural Health Monitoring (SHM). This study endeavors to introduce a simplified data-driven approach catering to System Identification (SI) and Response Estimation (RE). This is realized through the utilization of a summation of sine functions, fashioned as a model to harmonize with time domain vibration and acoustic responses. The fidelity of the findings is subsequently authenticated through the application of the Frequency Domain Decomposition (FDD) technique. In addition to the identification process, the proposed approach extends its applicability to predicting time domain responses at novel locations. This augmentation is achieved by harnessing an enhanced methodology founded on the principles of the Dynamic Mode Decomposition (DMD) technique. The veracity of these predicted outcomes is underscored through a comparison with measurements recorded at the same locations, alongside concurrent analysis of DMD-derived results. In order to affirm the efficacy of the proposed methodology, a case study involving a building grappling with enigmatic vibration issues is meticulously selected. The findings underscore that the proposed technique not only adeptly discerns unidentified vibrations without resorting to frequency domain transformation techniques, but also facilitates precise estimation of time domain responses.

Spectrum Extension of a Real-Aperture Microwave Radiometer Using a Spectrum Extension Convolutional Neural Network for Spatial Resolution Enhancement

Enhancing the spatial resolution of real-aperture microwave radiometers is an essential research topic. The accuracy of the numerical values of brightness temperatures (BTs) observed using microwave radiometers directly affects the precision of the retrieval of marine environmental parameters. Hence, ensuring the accuracy of the enhanced brightness temperature values is of paramount importance when striving to enhance spatial resolution. A spectrum extension (SE) method is proposed in this paper, which restores the suppressed high-frequency components in the scene BT spectrum through frequency domain transformation and calculations, specifically, dividing the observed BT spectrum by the conjugate of the antenna pattern spectrum and applying a Taylor approximation to suppress error amplification, thereby extending the observed BT spectrum. By using a convolutional neural network to correct errors in the calculated spectrum and then reconstructing the BT through inverse fast Fourier transform (IFFT), the enhanced BTs are obtained. Since the extended BT spectrum contains more high-frequency components, namely, the spectrum is closer to that of the original scene BT, the reconstructed BT not only achieves an enhancement in spatial resolution, but also an improvement in the accuracy of BT values. Both the results from simulated data and satellite-measured data processing illustrate that the SE method is able to enhance the spatial resolution of real-aperture microwave radiometers and concurrently improve the accuracy of BT values.

A 34.7 µW Speech Keyword Spotting IC Based on Subband Energy Feature Extraction

In the era of the Internet of Things (IoT), voice control has enhanced human–machine interaction and the accuracy of keyword spotting (KWS) algorithms has reached 97%; however, the high power consumption of KWS algorithms caused by their huge computing and storage requirements has limited their application in Artificial Intelligence of Things (AIoT) devices. In this study, voice features are extracted by utilizing the fast discrete cosine transform (FDCT) for frequency-domain transformation and to shorten the process of calculating the logarithmic spectrum and cepstrum. The designed KWS system is a two-stage wake-up system, with a sound detection (SD) awakening KWS. The inference process of the KWS network is achieved using time-division computation, reducing the KWS clock to an ultra-low frequency of 24 kHz.At the same time, the implementation of a depthwise separable convolution neural network (DSCNN) greatly reduces the parameter quantity and computation. Under the GSMC 0.11 µm technology, post-layout simulation results show that the total synthesized area of the entire system circuit is 0.58 mm2, the power consumption is 34.7 µW, and the F1-score of the KWS is 0.89 with 10 dB noise, which makes it suitable as a KWS system in AIoT devices.

Evaluations of Clarinet Sound Quality and Register Characteristics

The clarinet is a commonly used woodwind instrument that has a unique sound quality, making it a prominent instrument in solo performances and orchestral music. Therefore, sound quality plays an important role in the perception of clarinet sound. This study provides a review of the relevant research in the field of clarinet sound analysis and synthesis, and identifies many valuable references from previous studies. In studying the noise component of clarinet sound, this study recorded audio samples of E3, E4, and E5 and preprocessed them by slicing and performing frequency domain transformations. Using 13 representative spectral features, this paper extracted features from the spectral signals and constructed a classification model of sound quality based on the SVM algorithm of machine learning. Finally, this study analyzed the shortcomings of the SVM algorithm in high-pitch analysis and proposed future research directions of combining the SVM algorithm with different sound characteristics of clarinet registers and spectral changes throughout the entire process of playing. These results shed light on guiding further exploration of the timbre of the clarinet.

TinyNeRF: Towards 100 x Compression of Voxel Radiance Fields

Voxel grid representation of 3D scene properties has been widely used to improve the training or rendering speed of the Neural Radiance Fields (NeRF) while at the same time achieving high synthesis quality. However, these methods accelerate the original NeRF at the expense of extra storage demand, which hinders their applications in many scenarios. To solve this limitation, we present TinyNeRF, a three-stage pipeline: frequency domain transformation, pruning and quantization that work together to reduce the storage demand of the voxel grids with little to no effects on their speed and synthesis quality. Based on the prior knowledge of visual signals sparsity in the frequency domain, we convert the original voxel grids in the frequency domain via block-wise discrete cosine transformation (DCT). Next, we apply pruning and quantization to enforce the DCT coefficients to be sparse and low-bit. Our method can be optimized from scratch in an end-to-end manner, and can typically compress the original models by 2 orders of magnitude with minimal sacrifice on speed and synthesis quality.

Research on Abnormal Feature Extraction and Early Fault Alarm Method of Rolling Bearing's Based on CDAE and KLD

A rolling bearing is an important part of rotating machinery, and it is widely used in the petrochemical industry, aerospace industry and other industries. Hence, it is of great significance to carry out condition monitoring and fault alarms for rolling bearings. Aiming at the problem of the rolling bearing fault, a method of an improved deep convolutional denoising auto encoder abnormal feature extraction and the Kullback-Leibler divergence threshold alarm is proposed. The experiment verification is carried out on the rotor bearing experiment platform. The experiment results show that the proposed method has good denoising performance and micro fault feature extraction ability under the condition of no fault data training and no frequency domain transformation. High accuracy, good efficiency and strong robustness of the proposed method for an early fault alarm are demonstrated by the experiment as well.

Machine tool spindle vibration monitoring system based on Bluetooth wireless network and LabVIEW

The spindle system is important in the machine tool. The vibration signal generated by the spindle system in the machine tool has a great influence on the machining accuracy of the product and the abrasion and lifetime of the tool. Therefore, it is necessary to monitor the vibration signal in the machining process of the machine tool spindle in real time. The vibration sensor is used to monitor the vibration signal of the machine tool spindle in this paper. The data collected by the sensors are transmitted to the Bluetooth module by the serial port for wireless communication. The vibration signal and its frequency domain transformation of the spindle are received by LabVIEW software platform, and the original waveform is displayed, the threshold alarm and the storage of data are realized. The experimental results show that the designed system can realize the online monitoring of the vibration state of the machine tool spindle effectively.

Implementation and analysis of symmetrical signals in the sequency domain

Every transform has unique attributes and traits that are crucial to reducing computing costs and offering simple solutions. Many different frequency domain transformations, for instance, have properties that can be used in a variety of signal processing applications and analyses. Some of the Complex Hadamard Transform's variants' sequencies can be compared to those of the Discrete Fourier Transforms. It is proven the characteristics of the Conjugate Symmetric Sequency-Ordered Complex Hadamard Transform symmetry. These qualities are crucial for signal analysis and image processing. Due to duplicate spectra across the origin, it reduces computational complexity, makes an analytical analysis for symmetric signals simpler, and needs less storage. Its analysis shows that the Discrete Fourier Transform and this Complex Hadamard Transform version exhibit similar symmetry tendencies. By using elementary signals in the time domain to connect the positive and negative sequencies with their associated phasor conceptions, sequency domain spectra are used to highlight these properties. As a result of image representation, relative spectra are represented in related domains. Transform can be used to extract and analyze various aspects from a wide range of medical images.

Zero-Watermarking for Vector Maps Combining Spatial and Frequency Domain Based on Constrained Delaunay Triangulation Network and Discrete Fourier Transform

With its lossless properties, zero-watermarking has attracted a lot of attention in the field of copyright protection for vector maps. However, the common zero-watermarking algorithm puts too much emphasis on mining for global features, making it vulnerable to cropping attacks, and the robustness is not comprehensive enough. This study provides a vector map zero-watermarking scheme that utilizes spatial statistical information and frequency domain transformation methods in an effort to solve the aforementioned issue. In order to make the scheme more resistant to cropping and compression, it is constructed on the basis of feature point extraction and point constraint blocking of the original vector map. Within each sub-block, feature points are used to build constraint Delaunay triangulation networks (CDTN), and the angular values within the triangle networks are then extracted as spatial statistics. The angle value sequence is further transformed by discrete Fourier transform (DFT), and the binarized phase sequence is used as the final feature information to build a zero watermark by executing an exclusive disjunction operation with the encrypted copyright watermark image, both of which contribute to the scheme's robustness and security. The results of the attack experiments show that the proposed vector map zero-watermarking can restore identifiable copyright images under common geometric attacks, cropping attacks, and coordinate system transformations, demonstrating a high level of robustness. The theoretical basis for the robustness of this watermarking scheme is the stability of CDTN and the geometric invariance of DFT coefficients, and both theory and experiment validate the method's validity.

Real-time cable force identification based on block recursive Capon spectral estimation method

Cables are critical force-bearing components of cable-supported structures that greatly influence structure security. The changes in cable forces can reflect the cable service state and provide an important basis for structural health assessment. This paper proposes a method for real-time cable force identification (CFI) to tracking its varying rules. First, a recursive Capon method based on block updating is derived to estimate the amplitude spectrum at each observation moment. Second, determine the natural frequency (and corresponding mode order) in the transformed frequency domain and calculate the cable tension according to related formulas. Proper algorithm optimization and parameter selection are studied to ensure that the calculation cost meets the real-time requirement of CFI. The proposed method is applied to an experimental cable and the cables of a half-through arch bridge for validation.

Wave propagation in thin pretwisted composite strips with an embedded delamination

In this work, a thin pretwisted and delaminated composite strip is modeled using the variational asymptotic method (VAM). The original 3D problem is reduced to a 2D cross-sectional analysis and a 1D problem along the length of the strip. Damage, in the form of an embedded delamination, is accounted in the strip by following a sublaminate approach. The 1D governing elastodynamic equations of the strip model is solved using the spectral finite element (SFE) method, wherein the structure is modeled as a wave guide. In SFE method, the governing elastodynamic equations are solved in transformed frequency domain to obtain the dynamic stiffness matrix. The guided wave response of the delaminated composite structure for different loads and boundary conditions are subsequently computed. Modal and wave response from SFE using the reduced 1D model are validated by comparing results with those reported in the literature and in some cases with the results obtained from a 3D model using a direct finite element (FE) package. Modal and wave propagation analysis is performed for various dimensions of delamination and composite layup. The analysis provides insight into the dynamic characteristic of partially delaminated strip. The usage of SFE is expected to yield a faster result and lesser computation time than the conventional direct FE method for analyzing and monitoring the delaminated strip exposed to high frequency excitations. Additionally, the capability of the model to investigate the inverse problem of damage detection is demonstrated.

Transformations of borehole magnetic data in the frequency domain and estimation of the total magnetization direction: A case study from the Mengku iron-ore deposit, Northwest China

Borehole magnetic prospecting measures the three components of the magnetic field and is sensitive to the vertical depth of the magnetic source, which plays an important role in deep mineral exploration. Magnetic field transformations in the frequency domain constitute common and important processing for ground and airborne data but are rarely applied to borehole magnetic data. Here, we deduce component transformation and magnetization-direction transformation formulas for borehole data. It is found that the transformation factors differ from those of ground and airborne data. The new formulas depend on the position of the drillholes. The transformed data are then used to estimate the total magnetization direction by computing the correlation coefficient between the transformed components and the amplitude anomaly. The advantages of this method are that the borehole data yield a local and accurate magnetization direction and the total amplitude anomaly is computed directly from the individual anomalies in the observed components. The frequency-domain transformation is tested on the synthetic data, and the estimated directions are consistent with the true values. The processing procedure is applied to borehole data collected at the Mengku iron-ore deposit in Northwest China, where magnetic surveys from 55 boreholes have been acquired. The estimated magnetization directions yield good approximations of the results from the ground magnetic data and the physical property measurements of rock samples.