Search

Knee osteoarthritis (OA) is a prevalent musculoskeletal condition affecting millions worldwide, posing significant health and economic burdens. Characterized by the degeneration of joint cartilage, the progression of knee OA varies significantly among individuals, making its prediction a complex issue. Previous studies on automated knee OA diagnosis have primarily relied on unimodal data, often overlooking the valuable information present in multi-modal data. Multi-modal learning, which integrates information from various modalities, is increasingly recognized for its potential to enhance diagnostic performance in medical applications. However, such models incur a higher computational load due to the additional data required. This research investigates the feasibility of multi-modal neural networks in knee OA diagnosis by integrating structural demographic data with unstructured imaging data. Three deep learning unimodal models (InceptionV3, DIKO, and EfficientNetv2) were transformed into multi-modal architectures (MF_InceptionNet, MF_DIKO, and MF_Eff) to compare their diagnostic capabilities. The proposed multi-modal models share a common architecture, with unimodal models acting as image feature extraction backbones and separate embedding layers for demographic data. The image features and demographic embeddings are combined into a unified vector before classification. Extensive experiments were conducted to evaluate the performance of these models across different class categories and dataset sizes. MF_DIKO and InceptionV3 emerged as the best multi-modal and unimodal neural networks, respectively, with overall accuracies of 0.67 and 0.75 for 3-class severity classification. Contrary to existing literature, our findings reveal that unimodal neural networks using only imaging features outperform multi-modal networks, suggesting unimodal models might suffice in certain applications.

3D lane detection from the input monocular image is a basic but indispensable task in the environment perception of automatic driving. Recent work uses modules such as depth estimation, coordinate system transformation, and time series tracking to achieve the correspondence of 2D to 3D information. However, inaccurate depth information generated by perturbations during conversion poses a challenge to lane detection methods that rely only on monocular images. To solve the above problems, we propose a PELD model, a proxy transformation using BEV aerial view, to explicitly give 3D lane detection results. Specifically, when sampling feature information, feature flipping is proposed to supplement the global context information before view conversion, and the 3D position encoding information generated by the forward-looking features enhances the depth information. After the 3D position encoding information is combined with the feature information, the cross-attention module is used as a value for adaptive supervision of BEV queries. On the one hand, we use deformable attention to sample forward looking features and generate explicit lane representation; on the other hand, we supplement supervised lane line generation by supplementing forward looking features and enhancing 3D spatial information. PELD implements a more advanced approach than ever before on OpenLane and Apollo datasets.

Crowdsourcing platforms provide an efficient and cost-effective means to acquire the extensive labeled data necessary for supervised learning. However, the labels provided by untrained crowdsourcing workers often contain a considerable amount of noise. Although the application of ground truth inference algorithms to deduce integrated labels effectively enhances label quality, a certain level of noise persists. To further diminish the noise within crowdsourced labeling, this paper introduces a novel Small Loss-based Noise Correction algorithm (SLNC). SLNC first filters the crowdsourced data, leveraging the characteristic of neural networks to preferentially fits clean samples, thereby obtaining relatively clean and noisy sets. It then employs data augmentation techniques to enhance the clean set and subsequently trains the corrector on this augmented set to rectify the noisy set. SLNC has been evaluated using 16 simulated and two real-world datasets. The results indicate that SLNC surpasses comparative algorithms in the quality of the final labels.

Adaptive optics (AO) can effectively improve the beam quality of solid-state slab lasers. However, the aperture of the output beam increases as the output power of the laser increases, resulting in a larger measurement system. Ultimately, a more complex AO system needs to be designed. To meet the requirements of conjugate imaging in an AO system, it is of research significance to coordinate and optimize the system's structural dimensional parameters while enabling the detection of multiple parameters, such as the wave-front and beam quality. In this paper, a multiparameter measurement technique in a large-aperture rectangular laser beam aberration-correction system is proposed. The system layout is optimized with total dimensions of 400 mm × 150 mm × 246 mm (L × W × H). The AO system conforms to the requirements for conjugate detection and can perform wave-front detection, far-field evaluation, and near-field detection of a 160 mm × 120 mm rectangular beam produced from a solid-state slab laser. The findings reveal that the measured PV values of wave-fronts of the measurement system are less than 0.288 μm, the RMS is no more than 0.079 μm, the average far-field beam quality factor is 1.248 times the diffraction limit, and the average near-field beam uniformity is 0.533 at a temperature of 20 °C ± 10 °C; these results satisfy the technical parameters.

A laser time-frequency analysis ranging method based on wavelet transform for underwater target detection has been proposed in this paper. The band-pass filtering characteristics of the wavelet transform are combined with the energy conservation of the signal time-frequency domain during the wavelet time-frequency decomposition process of the echo signal. Binary spline interpolation is used to obtain the time-frequency correspondence matrix of the echo signal energy. In this method, the frequency range of the target echo signal is selected through the multi-layer wavelet time-domain decomposition process. The time-frequency energy matrix of the echo signal is extracted in the frequency range, and these frequencies are integrated to determine the restored time-domain echo signal. The corresponding delay of the target is determined by the peak detection method, and the target distance information is retrieved. The emission source is a 532 nm pulsed laser with a repetition frequency of 50 kHz, a pulse width of 1 ns, and average power of 3 W. The APD detector and the acquisition system with a bandwidth of 1 G are used to receive the echo signal. By detecting diffuse reflection targets in the water with an attenuation length of 10, the signal processing method can reduce the ranging error from 11.83 cm to less than 4 cm.

Accurate measurement of the spatial angle is key for plane flatness and guide rail straightness detection. This paper presents a lens-array-based optical system and an analytical model for micro-spatial angle measurements. In this system, a collimated light beam passes through four-array lenses arranged in a pyramid shape and forms regular array spots on a charge-coupled device sensor. The angles with respect to the $X$ and $Y$ axes are calculated by analyzing the distance of these spots on the CCD, the distance between adjacent apertures on the lens array, and the inclination angle between the lens array and the CCD. Using the coordinate value of the array spots on the CCD, the angle around the Z-axis can also be calculated simultaneously. Finally, the accuracy of the proposed method was verified by comparing the measured results with those of the autocollimator, and it was shown that the proposed approach enabled the achievement of RMS $\le 0.1''$ . Additionally, this measurement system which is consists of a laser source, lens arrays and CCD, compared with other methods, is smaller in volume and more convenient to carry.

Temperature compensation occupies a significant function in the insulation condition evaluation of converter transformer bushings based on the Frequency Domain Spectroscopy (FDS), which could influence the accuracy of its insulation diagnosis. In order to accurately obtain the variation law of FDS at different temperature, a Resin Impregnated Paper (RIP) insulation bushing for converter transformer was designed by the equivalence of maximum electric field, coaxial cylinder structure and RIP materials. The increment law of dielectric parameters both dissipation factor and capacitance at wideband frequency (1mHz ~ 10kHz) was measured for a quality RIP insulation samples at the temperature from 30°C to 100°C. 1) Under the uniform temperature, it was found that the value of dissipation factor was rising with the frequency from 10 Hz to 0.001Hz, while the tendency was reversed at the frequency from 100Hz to10kHz. 2) The minimum value of tan $\delta $ from 0.001Hz to 10kHz could be driven to higher frequency moving (10Hz to 10kHz) by the action of temperature rising 30°C to 100°C.3) The effect of temperature on the lower frequency 0.001Hz ~ 1Hz tan $\delta $ of RIP insulation closely conformed to the index law of Arrhenius equation, while it was in accordance with the linear decrease at higher frequency 1kHz ~ 10kHz. The effect of temperature on dissipation factor could be eliminated by curve segment shifting based the experiment equation, which it at 0.001Hz dissipation factor could be eliminated by curve shifting based the index equation originated Arrhenius equation. The method is beneficial to eliminate the influence of temperature on the dielectric spectroscopy test results, and then to eliminate the influence of temperature on the dielectric spectroscopy diagnosis of RIP bushings.

The simulated star map generated by the dynamic star simulator cannot meet the current demand for calibration of the star map recognition rate of the star sensor due to the huge amount of guide star catalog data and the slow retrieval process. This paper proposes the guide star equipartition method, which makes the constant number of each sub-region of the guide star catalog basically the same. This method first divides the declination at equal intervals through the field of view of the dynamic star simulator, and then determines the interval of the right ascension by the average number of stars with magnitude Mv that can be photographed in the field of view of the star sensor. The SAO (Smithsonian Astrophysical Observatory) catalog containing 5103 stars in the whole celestial sphere is divided into 452 sub-areas by the average number of guide stars. After the division, the guide star retrieval time is about 8ms, and the refresh rate of the simulated star map is increased by more than three times. The real-time requirements of the dynamic star simulator have been improved. Based on this, a star map simulation software is designed, which can realize the division of star catalogs with different fields of view and different magnitude thresholds.

We demonstrate a temperature-insensitive and high-sensitive label-free biosensor for low-concentration Human IgG detection based on S-tapered optical fiber (STF). The sensor is sensitive to local refractive index changes and has high robustness against temperature noise without a temperature control system. The temperature sensitivity of the sensor is −0.02nm/°C in the range of 25°C-45°C. Within 15 minutes at room temperature, the average drift of the center wavelength is 0.04 nm. The calibration experiments demonstrate that STFs for liquid achieved a sensing precision of 0.04 nm and a sensitivity of 538.62 nm/RIU. The STFs are applied for the concentration measurement of multiple diluted ex vivo IgG biomolecule solutions. The results showed that STFs were reliable for low-concentration IgG detection and its lower limit of measurement was demonstrated as 28 ng/ml. With the advantages of temperature-insensitive, compact size, low cost, high precision, real-time detection, fast response, good specificity, this STF sensors are promising for the antigen detection applications.

In this paper, a novel modulation and multiplexing technology for reducing out-of-band leakage is proposed for multicarrier schemes based on CP-OFDM. The proposed method is to spread data of N layers with N selected spreading codes, then multiplex them into a data sequence and transmit it on 2K subcarriers. The N spreading codes, which can suppress out-of-band leakage, are selected out from a complete set of mutual orthogonal spreading codes. The simulation results show that the data modulation and multiplexing technology effectively reduces interference with neighboring frequency bands or adjacent sub-bands for different numerologies and so significantly improves spectrum efficiency

Wirelessly Amplified NMR Detectors (WAND) can utilize wireless pumping power to amplify MRI signals in situ for sensitivity enhancement of deep-lying tissues that are difficult to access by conventional surface coils. To reconfigure between selective and simultaneous activation in a multielement array, each WAND has a dipole resonance mode for MR signal acquisition and two butterfly modes that support counter-rotating current circulation. Because detectors in the same row share the same lower butterfly frequency but different higher butterfly frequency, a pumping signal at the sum frequency of the dipole mode and the higher butterfly mode can selectively activate individual resonators, leading to 4-fold sensitivity gain over passive coupling. Meanwhile, a pumping signal at the sum frequency of the dipole mode and the lower butterfly mode can simultaneously activate multiple resonators in the same row, leading to 3-fold sensitivity gain over passive coupling. When multiple rows of detectors are parallelly aligned, each row has a unique lower butterfly frequency for consecutive activation during the acquisition interval of the others. This wireless detector array can be embedded beneath a headpost that is normally required for multi-modal brain imaging, enabling easy reconfiguration between focal imaging of individual vessels and multiregional mapping of brain connectivity.

Wireless transmission of locally detected RF signals is necessary for long-term operation of batteryless and embedded transducers. To improve signal transmission efficiency over larger distances, multi-stage circuits were employed to down-convert RF signals before encoding them onto the emitted carrier wave. Such multi-stage arrangement had complicated design and high-power consumption. Here, a compact and low-power wireless modulator is introduced to directly encode input RF signals onto its oscillation carrier wave. The modulator consists of a double frequency parametric resonator that is overlaid with a single frequency passive resonator to create three resonance modes. By properly adjusting the substrate thickness between resonators, the highest resonance frequency is tuned to approximately the sum of lower two resonance frequencies, enabling efficient conversion of wireless pumping power into sustained oscillation currents. When an input RF signal is present with a certain frequency offset, the oscillation signal can be frequency modulated by the input signal to create multiple modulation sidebands separated by the offset frequency. The frequency encoded carrier wave can transmit MRI signals over larger distance separations to maintain constant image sensitivity, making the modulator useful to improve the remote detectability of miniaturized implantable and interventional devices.

We propose dual un-compressed video watermarking algorithms based on human visual system and scale invariant feature transform (SIFT) in the contourlet domain. First, every frame is decomposed by contourlet transform to obtain the low-frequency band and several high-frequency sub-bands in different directions. The energy value of every high-frequency sub-band is calculated. The low-frequency sub-band and the high-frequency sub-bands of the largest energy values are chosen to be embedded into watermarking signals. Second, the first watermarking algorithm is realized to embed watermarking signals by adjusting the size ratio between the coefficient histograms of low-frequency sub-band because the coefficient histogram of low-frequency sub-band possesses invariance to conventional geometric transform. Third, to the second watermarking algorithm, the human visual masking threshold based on contourlet domain is calculated as the maximum embedding tolerance, which is embedded in the largest energy value of the high-frequency sub-band as watermarking signals to guarantee the imperceptibility. Finally, when the watermarked video frame suffered from geometrically attacks, the SIFT and fine corrections are combined to effectively correct and restore the synchronization between geometrically attacked video frames and watermarking signals. The experimental results indicate that: 1) the visual masking based on video motion information in the contourlet domain can effectively improve the transparency and 2) combining the histogram mapping in low-frequency sub-band coefficients with SIFT and fine corrections can resist common geometric attacks and therefore improve the robustness of the watermarking algorithm.

Fault characteristic frequency is the main basis for rolling element bearing diagnostics but finding a suitable frequency band for demodulation and searching for the fault characteristic frequencies consume a lot of time and manpower in practice. A data-driven method based on recursive variational mode decomposition (RVMD), and an envelope order capture is proposed to realize the automatic fault diagnosis of bearing under different operating conditions. The process starts with a new proposed RVMD of the vibration signal, where the mode with maximum kurtosis of the unbiased autocorrelation of the envelope is selected to get envelope order spectrum. Thereafter, an order capture algorithm is designed to automatically search for the fault characteristic orders in theory, which are used for constructing feature vectors for diagnosis. The proposed method is tested on two test-beds which both contain the same type of bearing (SKF6205) but operate in different conditions, and gets good performance in bearing diagnosis. In addition, the fault diagnosis of test-bed two using training samples that are from test-bed one is investigated. This method reveals well generalization capability in the fault diagnosis of the same type of rolling element bearing under different operating conditions.

In this study, red, green, cyan, warm white and cool white (RGCWW) LEDs are individually controlled to simulate dynamic daylight. The spectral power distributions (SPDs) are measured for both the daylight and the mixed white light. The correlated color temperature (CCT), circadian action factor (CAF) curves in the daytime of the mixed white LED almost overlap to those of daylight with a small error of about 1%. The CIE chromaticity coordinates of the mixed white light are distributed along the Planckian locus with a little deviation below 0.003. Also, its color rendering index (CRI) is above 90. The RGCWW five-chip LEDs of high circadian and visual performances will benefit to synchronize circadian rhythm in closed environment during long time without daylight, combined with better mood and alertness. The simulation errors and development trend of the mixed lighting are discussed.

Traditional biological neural networks lack the capability of reflecting variable synaptic weights when simulating associative memory of human brains. In this paper, we propose a novel memristive multidirectional associative memory neural networks (MAMNNs) model with mixed time-varying delays. More precisely, the proposed model is investigated with time-varying delays and distributed delays. Then, we design two kinds of delay-independent and delay-dependent controllers to analyze the problem of finite-time synchronization. Based on the drive-response concept and Lyapunov function, some sufficient criteria guaranteeing the finite-time synchronization of the drive-response system are derived. With the removal of certain constraints on the weight parameters, the results we obtained for synchronization are less conservative. To illustrate the chaotic characteristics of the memristive MAMNNs, an image encryption scheme is designed. Meanwhile, the effectiveness of the proposed theories is validated with numerical experiments.

In this paper, we investigate the synchronization in the mean square sense of memristive multidirectional associative memory neural networks with mixed time-varying delays and stochastic perturbations. In the proposed approach, the mixed delays include time-varying delays and distributed time delays. Sufficient criteria guaranteeing the synchronization of the drive-response system are derived based on the drive-response concept, the stochastic differential theory and Lyapunov function. With the removal of certain strict conditions of weight parameters, less conservative results are generated. To illustrate the performance of the proposed synchronization criteria, a secure communication scheme to realize secure data transmission is designed. Meanwhile, the effectiveness of the proposed theories is validated with numerical experiments.