Search

A strong and reliable Traffic Sign Detection and Recognition (TSDR) system is essential for the effective deployment of autonomous driving technology. In this field, numerous scholars have conducted extensive research, but current studies only consider TSDR under ideal conditions, neglecting scenarios such as rain, snow, fog, which can cause image blurring. This paper investigates the challenges of TSDR performance degradation caused by five adverse environmental conditions: rain, snow, fog, lens dirt, and lens blur. To overcome the adverse effects of these conditions on TSDR, this paper proposes a Convolutional Neural Network (CNN)-based TSDR method, consisting of three modules: adverse environment classification module, image enhancement module, and traffic sign detection module. The adverse environment classifier, based on the VGG19 architecture, identifies whether the image includes the aforementioned five adverse weather conditions. The image enhancement module named Enhance-Net enhances each of the five adverse environments separately and specifically enhances the traffic sign regions within the image, rather than the entire image area. To increase the speed of the proposed method, the traffic sign detection module utilizes the YOLOv4 framework. The proposed method’s effectiveness is assessed using the CURE-TSD dataset, which includes traffic videos recorded in various adverse environmental conditions. Experimental results demonstrate that under five different levels of adverse environments, the pro-posed method achieves 95.03% accuracy and runs at a rate of 12.79 fps (frames per second). In contrast to the current benchmark, although there is a 2.81% reduction in accuracy resulting from training the proposed method on a subset of the dataset, the speed has increased by 12.03 fps, demonstrating the efficacy of the proposed approach.

Predicting the price of soybean meal futures has always been a hot topic in the financial market. It is influenced by many factors and has complex nonlinear characteristics, which makes it difficult to predict accurately. This paper proposes a prediction model for soybean meal futures price based on CNN-ILSTM-SA, which consists of CNN (Convolutional Neural Network), ILSTM (Improved Long Short-Term Memory), and SA (Self-Attention mechanism). CNN can well perform feature extraction on the input time series data. The ILSTM model presented in this paper removes the output gate of LSTM (Long Short-Term Memory) and improves the forget gate and input gate, thus reducing ILSTM’s complexity, shortening training time, and improving learning ability. SA can get the correlation between different input terms and predicted values, and assign different weight coefficients according to the correlation level to improve prediction accuracy. To verify the CNN-ILSTM-SA’s effectiveness, we compare it with eighteen baseline models, and the experimental results show that it is optimal. The MAE (Mean Absolute Error) of CNN-ILSTM-SA is 39.0441, the smallest of all models. As the data of the soybean meal futures price fluctuates in the range of 3500-4500 in experiments, the proportion of the MAE value of 39.0441 is 0.87%-1.15%, which is at an extremely low level; $\mathrm {R}^{2}$ (R-square) is 0.97487, which is closest to 1; The training time is 157.64s.

The rapid development of the Internet of Things (IoT) has put great challenges on the next generation's communication technologies. Optical Camera Communication (OCC), with its advantages of the green principle, less electromagnetic interference, and harmonious integrations, can serve as a complementary, and sometimes viable alternative technology to mainstream Radio Frequency (RF)-based one. Designing efficient OCC platforms crucially involves accurate modeling of the OCC system and evaluating the efficiency of OCC-based techniques, which then stimulates related practical applications. This paper aims to provide a systematic perspective to understand OCC and to present design guidelines for OCC systems from theory to practice. First, the main parameters and analytical models for OCC basic components are presented to provide guidance on hardware parameter selection and system simulations. Next, typical implementation procedures and common performance evaluation criteria for an overall OCC system are provided. Then, detailed goal-oriented OCC enhancement solutions are summarized aiming to provide systematic guidelines for system performance enhancement. Finally, OCC designed examples are presented based on the results of our OCC experimental platform and some future research directions are discussed.

Movement disorder patients are usually unable to easily communicate with the outside world. Therefore, a novel approach for EOG character input based on DC-CNN is proposed in this paper. Users only need to control the eyeball movement and eye blinking so as to manipulate the interactive interface on the screen for character input. The signal acquisition device used in this study adopts a dual-channel acquisition method. The EOG signals are preprocessed, converted to digital signals, and filtered for noise using bandpass filtering and wavelet transform. The filtered and denoised signals are input into the proposed novel approach for training the classification model. During the real-time interaction, the EOG signals are input into the trained model for classification. Next, the classification result is used as the input signal for a virtual keyboard which is compiled by PYQT5. After that, the virtual keyboard can generate corresponding responses depending on different input signals to achieve the character input. The model achieved a high classification accuracy of 99.3% during training and 98.5% in real-time classification, significantly outperforming existing methods. Furthermore, the accuracy of single character input and paragraph input can arrive at 98.2% and 97.6% as well, respectively. This novel DC-CNN approach offers a reliable and user-friendly communication method for movement disorder patients, enhancing their ability to interact with the world.

Building information serves as a critical foundational dataset in the fields of urban planning, smart cities and surveying and mapping, and high-resolution remote sensing (HRRS) imagery has become a vital data source for extracting building information. However, automatically extracting building information from HRRS imagery using a single feature or method remains a challenging task. On one hand, buildings exhibit significant variations in terms of their size, color, geometric structures, and other aspects. On the other hand, there are also numerous features in the environment that bear spectral and morphological resemblances to buildings. In this article, we proposed a multifeature fusion framework based on Dempster–Shafer (D-S) theory that consists of two steps for automated building extraction from HRRS imagery. The initial D-S fusion step involves two branches: object-level and pixel-level feature fusion. Then the outcomes are further combined to derive the ultimate building confidence information. In the framework, we introduced a proportional consistency and centroid consistency index to convert pixel-level features to object-level features, thereby facilitating their fusion. In addition, we proposed an initialization module for the basic probability assignment formula, enabling the elimination of the impact of nonbuilding objects and simplifying the construction process of BPAF. The experimental results based on the Nanjing, WHU, and Washington datasets demonstrate the effectiveness of our method, the accuracy outperforms the other four advanced algorithms.

P-band radiometry has been demonstrated to have a deeper sensing depth than L-band, making the consideration of multilayer microwave interactions necessary. In addition, the scattering and phase interference effects are different at the P-band, requiring a reconsideration of the need for coherent models. However, the impact remains to be clarified, and understanding the validity and limitations of these models at both L- and P-bands is crucial for their refinement and application. Therefore, two general categories of microwave emission models, including two stratified coherent models (Njoku and Wilhite) and four incoherent models (conventional tau-omega model and three multilayer models being zero-order, first-order, and incoherent solution), were intercompared for the first time on the same dataset. This evaluation utilized observations of L- and P-bands radiometry under different land cover conditions from a tower-based experiment in Victoria, Australia. Model estimations of brightness temperature (TB) were consistent with measurements, with the lowest root mean square error (RMSE) at P-band V-polarization under corn (2 K) and the highest RMSE at L-band H-polarization under bare soil (13 K). Coherent models performed slightly better than incoherent models under bare soil (3 K less RMSE), while the opposite was true under vegetated soil conditions (1 K less RMSE). Coherent and incoherent models showed maximum differences (3 K at P-band and 2 K at L-band), correlating strongly with soil moisture variations at 0–10 cm. Findings suggest that coherent and incoherent models performed similarly; thus, incoherent models may be preferable for estimating TB at L- and P-bands due to reduced computational complexity.

As a non-invasive neuromodulation technique, transcranial magnetic stimulation (TMS) has shown great potential in scientific research and clinical application. Using TMS to stimulate the deep brain is important for enhancing the therapeutic effects of mental disorders and explaining the causes of mental disorders. However, the intracranial induced electric field (E-field) generated by conventional magnetic stimulation coils dissipates severely with increasing stimulation depth and the focalized stimulation is limited to superficial areas. In this paper, we first propose a novel multi-quadrant spatial array (MQS array) based on curved-shaped coils. Each coil in the MQS array is tangent to the head and bent away from the human head, which is conducive to reducing the accumulation of non-longitudinal induced E-field components and improving the stimulation focalization in the deep brain. Then, we propose a new spatial magnetic array optimization method based on the BP-NSGA-II algorithm. The predictive models of deep brain transcranial magnetic stimulation characteristics of the spatial array are obtained by the BP neural network and the multi-objective optimization of the stimulation currents applied in the spatial array is performed with the NSGA-II algorithm. Results show that the MQS array can produce an obvious focusing area at 10 cm below the scalp which satisfies the depth requirement of deep brain transcranial magnetic stimulation. Under the same constraint of Joule loss, the optimized MQS array can enhance intracranial stimulation intensity by 131%, increase the longitudinal attenuation ratio to 4.5 times, and reduce the focusing area by 76% compared to the conventional planar magnetic stimulation array. The proposed MQS array has significant advantages in deep brain focalized stimulation, and the spatial array optimization method described in this study may provide a valuable reference for coil or array optimization processes in other application contexts.

Breast cancer is a type of disease that primarily affects the breast tissue, and it is crucial to achieve early diagnosis for successful treatment and recovery. In recent years, the residual network (ResNet) has gained significant attention in the detection of breast cancer using medical images. In this paper, we propose an efficient and robust deep learning framework called FastLeakyResNet-CIR, an improved ResNet architecture, for breast cancer detection and classification. The FastLeakyResNet-CIR achieves an impressive accuracy of 98.94% when evaluated on a dataset of 7909 microscopic images of breast tumor tissue from BreakHis dataset, which outperforms the state-of-the-art methods, e.g. ResNet18, ResNet50, InceptionV3 and VGG16. The experiment results further highlight the potential of FastLeakyResNet-CIR for accurate and rapid diagnosis of breast cancer, thus facilitating effective medical treatment for patients.

The trade-off between decoding performance and hardware costs has been a long-standing challenge in Low-Density Parity Check (LDPC) decoding. Based on model-driven methodology, the Neural Network-Aided Variable Weight Min-Sum (NN-aided vwMS) algorithm is proposed to address this dilemma in this paper. Not only eliminating the second minimum value in the check node update process for reducing hardware complexity, our approach featuring a fast-convergent shuffled scheduling method proposed to enhance convergence speed can also maintain similar decoding performance as compared to the traditional normalized min-sum algorithm. Different from existing model-driven methodologies only suitable for short codes, a Globally-Coupled Like (GC-like) LDPC code construction is presented to enable efficient training with simplified neural networks for longer LDPC codes. To demonstrate the capability of the NN-aided vwMS algorithm with the fast-convergent shuffled scheduling method, a GC-like (9126,8197) LDPC decoder is implemented for NAND flash applications, achieving a 6.56 Gbps throughput with a core area of $0.58~mm^{2}$ under the 40-nm CMOS TSMC process, and average power consumption of 288 mW under the frame error rate of $2.64 \times 10^{-5}$ at 4.5dB. Our decoder architecture achieves a superior normalized throughput-to-area ratio of $11.31~Gbps/mm^{2}$ , demonstrating a 2.4x improvement among previous works.

Based on Automatic Identification System (AIS) data in maritime settings, this paper explores the limitations of traditional Recurrent Neural Networks in extracting features from complex vessel trajectory sequences. We propose a fusion model, namely CNN-RGRU-Attention, for vessel trajectory prediction. The model integrates Convolutional Neural Network (CNN), Attention Mechanism, and Gated Recurrent Unit (GRU). The effectiveness of the model is validated using actual AIS data, demonstrating significant improvements in metrics such as Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE). The CNN-RGRU-Attention model provides crucial theoretical support for enhancing the safety management of maritime traffic services.

Synthetic aperture radar automatic target recognition (SAR-ATR) models based on deep neural networks (DNNs) are vulnerable to attacks of adversarial examples. Universal adversarial attack algorithms can help evaluate and improve the robustness of the SAR-ATR models and have become a research hotspot. However, current universal adversarial attack algorithms have limitations. First, considering the difficulty in obtaining information on the attacking SAR-ATR models, there is an urgent need to design a universal adversarial attack algorithm under a black-box scenario. Second, given the difficulty of acquiring synthetic aperture radar images, the effectiveness of attacks under small-sample conditions requires improvement. To address these limitations, this study proposed a black-box universal adversarial attack algorithm: transferable universal adversarial network (TUAN). Based on the idea of the generative adversarial network, we implemented the game of generator and attenuator to improve the transferability of universal adversarial perturbation (UAP). We designed loss functions for the generator and the attenuator, respectively, which can effectively improve the success rate of black-box attacks and the stealthiness of attacks. In addition, U-Net was used as a network structure of the generator and the attenuator to fully learn the distribution of examples, thereby enhancing the attack success rate under small-sample conditions. The TUAN attained a higher black-box attack success rate and superior stealthiness than up-to-date UAP algorithms in non-targeted and targeted attacks.

The fractured-vuggy carbonate reservoir comprises various types of storage and seepage spaces, and is composed of multi-scale dissolution pores and fractures. The frequent changes to working systems make the characteristics of water breakthrough complex, and the production data nonlinear and non-stationary, resulting in great difficulty in real-time prediction. Traditional production forecasting methods only consider temporal correlations, neglecting the spatial correlations between production wells and local geological features. In this paper, adopts a modular design approach that comprehensively considers the spatiotemporal characteristics by abstracting each production well in the unit as a directed graph network node. We establish a graph attention network module based on the connectivity between wells to simulate fluid motion patterns and extract spatial features. To address the autocorrelation characteristics of the production sequences, we use a self-attention mechanism module to capture the temporal dependency relationships between production sequences. Finally, considering the fusion of spatiotemporal features, a gating mechanism is designed to adaptively aggregate spatiotemporal characteristics produced by the previous two modules, enabling dynamic production forecasting. We validate our proposed model using real-world production data from the Tarim Basin in China. Our experimental results demonstrate the superiority of the new model over existing production prediction models in fractured-vuggy carbonate reservoirs.

Nowadays antiretroviral therapy is widely used to cut down the viral burden and cut off a few infection sources in clinical medicine for Human Immunodeficiency Virus (HIV) infected patients, and a lot of evidence have shown that a massive coverage with antiretroviral therapy has already acquired a series of contributions and successes to save life and popularize preventive knowledge. Based on the HIV dynamic model, this paper designs a feedback controller to adjust the concentration values of CD4+T cell, CD8+T cell and viral load in HIV dynamic model to a healthy condition asymptotically. The HIV dynamic model is expressed in the form of a polynomial system and then the control law, as the appropriate drug dosage usually applied in the medical therapy, is designed to suppress the reproduction of the HIV virus and improve the production rate of healthy cell in vivo, simultaneously. According to the polynomial Lyapunov theory and sum of squares (SOS) technique, the conditions for the controller synthesis are proposed. The simulation experimental results can be testified the availability of the proposed method.

With the rapid development of Earth observation technology and deep learning, building extraction from remotely sensed imagery based on deep convolutional neural networks has attracted wide attention in recent years. However, due to the heterogeneity of building shapes and sizes and the complexity of the surrounding objects, current building extraction methods still have challenges in boundary accuracy and complete building extraction. For these purposes, we proposed a low-level feature enhancement and multiscale attention pyramid aggregation network (LFEMAP-Net) that considers building boundary information and multiscale feature expression to obtain higher accuracy building extraction. First, a low-level feature enhancement model is proposed based on the prior edge information to enhance the representation of spatial details, effectively addressing issues related to information loss and fuzzy boundaries. Additionally, a multiscale attention pyramid aggregation model is developed during the decoding stage to facilitate the fusion of features from different scales, thereby enhancing the extraction of building features. The experimental results on two publicly available datasets validate that LFEMAP-Net can overcome building extraction interruptions and boundary blur in complex scenes, and achieve boundary optimization and complete segmentation of buildings and achieve better performance than other advanced semantic segmentation models.

Detection based on LiDAR and camera fusion is increasingly popular for researchers in the autonomous driving domain. Compared to the camera-only and LiDAR-only methods, the fusion-based methods indeed improve the detection accuracy on public-available datasets. However, due to the complexity of the projection or fusion mechanism, few of these methods can run in real time even on an advanced desktop GPU. Thus, in this paper, we propose a new fusion detection model G-Fusion with a light and fast image view-transform module. According to our receptive field analysis of image feature maps, we directly project image features to only one voxel layer located on the ground, then fuse the LiDAR and image features by concatenation and convolution. With this delicately designed module, G-Fusion greatly boosts the state-of-the-art speed performance on the nuScenes dataset, achieving a good balance with the competitive detection scores. Meanwhile, since the precision of sensor extrinsic parameters is important for most fusion-based methods, we also deeply dig into our model’s calibration error tolerance ability and discover the failure noise condition.

Traditional agricultural practices of hand-picking ripe tomatoes are labor-intensive and inefficient for large-scale harvesting. To address this, we propose an innovative approach using the YOLOv7 algorithm for ripe tomato detection, enabling robotic arms to perform the picking. However, the occlusion of tomatoes in the field often leads to unclear target features, causing false or missed detections. So it is worth studying and this paper proposes a tomato detection method based on improved YOLOv7. The novelty is shown below. First, a new structure called ReplkDext is redesigned to increase the receptive field. ReplkDext is introduced before the last layer of CBS in the backbone. Secondly, to overcome the problem of low FLOPS caused by frequent access to memory in traditional neural networks, the head structure of YOLOv7 is redesigned. By using FasterNet to optimize the structure between Concat and CBS in the head, FasterNet makes the model balance between running speed and detection accuracy. Finally, to improve the ability of convolution, ODConv is added after the last ELANN-2 structure in the Head layer. ODConv improves the feature extraction ability of small targets and obtains more feature information about ripe tomatoes. Experiments show that compared with YOLOv7, Map@.5 of Tomato-YOLOv7 has increased by 1.3%. The model is overall better than other models. The overall contribution of the Tomato-YOLO model is to provide important insights into agricultural product detection and provide a theoretical basis for automated tomato harvesting in orchards.

Multipolarimetric synthetic aperture radar (SAR) interferometric phase optimization and phase series estimation have received a lot of attentions recently from the polarimetry SAR interferometry (PolInSAR) community. In this article, a maximum likelihood estimation (MLE) method for the interferometric coherence matrix (ICM) is proposed, which is further applied to both interferometric phase optimization and phase series estimation. By modeling the PolInSAR coherence matrix as the Kronecker product of the polarimetric coherence matrix and ICM, the MLE of ICM under complex circular Gaussian distribution hypothesis is acquired through an alternate iterative optimization method. In addition, it is theoretically proved in this article that the two state-of-the-art methods, i.e., the TP (total power) method and the MLE-MPPL method, are suboptimal compared to the proposed method regarding the MLE of ICM. Numerical experiments are conducted on simulated fully polarimetric data, airborne fully polarimetric E-SAR data, and spaceborne dual polarimetric Sentinel-1A data, to confirm the effectiveness and superiority of the proposed method.

Cognitive impairment is typically reflected in the time and frequency variations of electroencephalography (EEG). Integrating time-domain and frequency-domain analysis methods is essential to better understand and assess cognitive ability. Timely identification of cognitive levels in early Parkinson’s disease (ePD) patients can help mitigate the risk of future dementia. For the investigation of the brain activity and states related to cognitive levels, this study recruited forty ePD patients for EEG microstate analysis, including 13 with mild cognitive impairment (MCI) and 27 without MCI (control group). To determine the specific frequency band on which the microstate analysis relies, a deep learning framework was employed to discern the frequency dependence of the cognitive level in ePD patients. The input to the convolutional neural network consisted of the power spectral density of multi-channel multi-point EEG signals. The visualization technique of gradient-weighted class activation mapping was utilized to extract the optimal frequency band for identifying MCI samples. Within this frequency band, microstate analysis was conducted and correlated with the Montreal Cognitive Assessment (MoCA) Scale. The deep neural network revealed significant differences in the 1-11.5Hz spectrum of the ePD-MCI group compared to the control group. In this characteristic frequency band, ePD-MCI patients exhibited a pattern of global microstate disorder. The coverage rate and occurrence frequency of microstate A and D increased significantly and were both negatively correlated with the MoCA scale. Meanwhile, the coverage, frequency and duration of microstate C decreased significantly and were positively correlated with the MoCA scale. Our work unveils abnormal microstate characteristics in ePD-MCI based on time-frequency fusion, enhancing our understanding of cognitively related brain dynamics and providing electrophysiological markers for ePD-MCI recognition.

Manipulation of light-matter interactions has great potential applications in optical modulators and optical switches. However, the strong coupling of anapole-cavity modes has not been involved yet. Here, we firstly propose a simple Si-based metasurface hybrid system to realize tunable strong coupling of anapole-Fabry-Pérot (FP) cavity via inserting a Si square nanocubes metasurface inside an FP nanocavity. A remarkable spectral splitting behavior is achieved with large Rabi splitting of 192 meV due to the strong resonant coupling between the anapole and FP cavity modes. The induced two hybrid energy states are respectively dominated by the anapole mode and cavity mode and can be tuned by the length of FP nanocavity and the thickness of SiO2 spacer, respectively. This Si-based metasurface system opens an avenue for designing optoelectronic devices that can manipulate light-matter interactions at nanoscales.

The characteristics of hydrological data include nonconsistency and nonlinearity. The prediction accuracy can be improved through the combination of both the decomposition algorithm and the runoff model. Previous studies have typically focused on the combination of a single decomposition algorithm and model. These studies have compared the prediction accuracy before and after decomposition, ignoring the role of multiple decomposition algorithms and models. Considering the limitations of previous single combinations of decomposition algorithms and models, this study will explore the unique features of hydrological data by using a combination of five algorithms, including Empirical Mode Decomposition (EMD), Ensemble Empirical Mode Decomposition (EEMD), TIME series decomposition (TIME), Variational Mode Decomposition (VMD), and Singular Spectrum Analysis (SSA). The study constructed models for Prophet, Long Short-Term Memory (LSTM), Multiple Regression (MLR), Random Forest Regression (RFR), Gradient Boosting Regression (GBR), and Support Vector Regression (SVR). Thirty combined prediction models were then developed and used to forecast medium and long-term runoff at Xianyang Station. To comprehensively evaluate the forecasted runoff results, multiple evaluation metrics were used. The prediction accuracy improved after using EMD and TIME decomposition, but the difference was insignificant, and TIME decomposition was the least effective. VMD, EEMD, and SSA, on the other hand, yielded higher data quality. The combined model achieved an NSE above 0.70, demonstrating good prediction results. Of the thirty combined models, the SSA-SVR and SSA-LSTM models were most accurate, with a verification NSE of 0.90. This study developed a comprehensive, reliable, and accurate combination prediction model by employing multiple decomposition algorithms and models. These findings provide a framework for characteristics-driven watershed runoff prediction and water resources scheduling.

Model-Based Reinforcement Learning (MBRL) has been gradually applied in the field of Robot Learning due to its excellent sample efficiency and asymptotic performance. However, for high-dimensional learning tasks in complex scenes, the exploration and stable training capabilities of the robot still need enhancement. In light of policy planning and policy optimization, we propose a bidirectional model-based policy optimization algorithm based on adaptive gaussian noise and improved confidence weights (BMPO-NW). The algorithm parameterizes bidirectional policy networks into noise networks by adding different adaptive Gaussian noises to the connection weights and biases. This can improve the randomness of policy search and induce efficient exploration for the robot. Simultaneously, the confidence weight of improved activation function is introduced into the Q-function update formula of SAC, which can reduce the error propagation problem of target Q-network, and enhance the robot’s training stability. Finally, we implement the improved algorithm based on the framework of bidirectional model-based policy optimization algorithm (BMPO) to ensure asymptotic performance and sample efficiency. Experimental results in MuJoCo benchmark environments demonstrate that the learning speed of BMPO-NW is about 20% higher than baseline methods, the average reward is about 15% higher than other MBRL methods, and 50%-70% higher than MFRL methods, while the training process is more stable. Ablation experiments and different variant design experiments further verify the feasibility and robustness. The research results provide theoretical support for the conclusion of this paper and hold significant practical value for MBRL to help the robot realize applications in complex scenarios.

Noise-shaping (NS) successive approximation register (SAR) analog-to-digital converters (ADCs) using passive integrators have drawn increasing attention owing to their simplicity and power efficiency. However, a capacitor array with a passive integrator result in a huge number of unit capacitors and power consumption. This paper presents a second-order fully passive NS-SAR ADC with a segmented capacitor array and a hybrid switching procedure. We use a most-significant-bit(MSB) sampling-segmented capacitor array to reduce the number of unit capacitors, and a common mode voltage(VCM)-based hybrid switching procedure to reduce the average switching power. Compared to fully passive NS-SAR ADCs with conventional capacitor array and switching procedure, the total number of unit capacitors is reduced by about 84%, and the average switching power is also reduced by about 97.8%. The prototype was fabricated using 180nm complementary metal oxide semiconductor(CMOS) technology, it only uses 478 unit capacitors, resulting in an area of 0.086 mm2. It consumes $16.5~\mu \text{W}$ , achieves 73.7 dB peak signal-to-noise-and-distortion ratio(SNDR), and a Schreier figure of merit(FoMs) of 165.5 dB.

Manipulation of light-matter interaction is critical in modern physics, especially in the strong coupling regime. Based on the highly localized field enhancement in the near field, we demonstrate that strong coupling can be achieved in the dielectric-metallic hybrid structure consisting of a perforated Si nanoblock filled with a slotted Ag nanodisk (nanoprism) in its center. Large Rabi splitting of 275 (262) meV is achieved due to the generation of new hybrid energy states via the strong anapole-plasmon coupling. Our results pave a feasible way to realize large Rabi splitting and offer an attractive platform to explore strong coupling regime in the light-matter interaction.

The prevalence of convolutional neural networks is hindered by their intricate structure and voluminous parameters, which consume significant processing resources during both training and inference. This study proposes a novel approach that involves treating convolutional kernels as tensors, utilizing suitable priori metrics to gauge their effectiveness, and employing a clustering algorithm to eliminate redundant convolutional kernels for network pruning. To measure the convolutional kernels, we use appropriate priori metrics and density-Based spatial clustering of applications with noise (DBSCAN) to cluster the convolutional kernels so that the cluster centroids are the reserved convolution kernels. The experimental results show that the method can effectively perform network lightweight while maintaining high accuracy. At the same time, the method can improve the efficiency of convolutional kernel utilization, thus reducing the computational resource consumption of the model. Empirical analyses conducted on datasets indicate that, in certain instances, the proposed pruning method outperforms established state-of-the-art methods.

With the continuous development of infrared technology, the application of infrared imagery is increasingly widespread. Nonetheless, infrared imagery suffers from low contrast and high noise characteristics, making it challenging to detect and recognize infrared small targets. Frequently, existing deep learning-based infrared target detection methods solely employ spatial information, neglecting frequency domain information, thereby creating susceptibility to background clutter interference and low detection accuracy in complex scenes. Here, a deep learning-based method is proposed that combines frequency domain clutter suppression and spatial feature extraction to detect infrared small targets. Firstly, the spatial domain infrared image is transformed into the frequency domain using block discrete cosine transform (DCT). Secondly, the frequency domain information is filtered using an attention mechanism to adaptively enhance the targets and suppress the background clutter. Finally, the compressed attention map is transformed back to the spatial domain and inputted into a lightweight neural network for spatial feature extraction and fusion to obtain detection results. This approach enhances detection accuracy, and reduces computational complexity and storage space, thereby making it practical and deployable. Experimental outcomes demonstrate strong detection capabilities on public datasets, and significant advantages over traditional and existing deep learning methods.

Unknown faults may occur in practical applications, necessitating an open-set classifier that can classify known classes as well as recognize unknown faults. The current deep open-set classification methods are implicit in optimizing the intra- or inter-class distances, which may result in performance degradation when the number of unknown classes far exceeds that of the known. In this study, the discriminative angle deep features for vibration signals are investigated. A novel normalized one-versus-all classification loss with center and contrastive regularization is proposed. The trained network can explicitly optimize deep features to ensure intra-class compactness and inter-class divergence. In this case, such discriminative features can be used for open-set fault classification. Furthermore, the effectiveness of the proposed method is verified using field-measured motor bearing and gear vibration signals. The results demonstrate the evident advantages of our proposed method over other approaches in practical fault diagnosis scenarios.

Multi-view partial multi-label learning (MVPML) is a fundenmental problem where each sample is linked to multiple kinds of features and candidate labels, including ground-truth and noise labels. The key problem of MVPML is how to manipulate the multiple features and recover the ground-truth labels from candidate label set. To this end, this study designs a novel Graph-based Multi-view Partial Multi-label model named as GMPM, which combines the multi-view information detection, valuable label selection and multi-label predictor model learning into a unified optimization model. To be specific, GMPM first exploits the consensus information across multiple views by learning the view-specific similarity graph and fuses multiple graphs into a target one. Then, we divide the observed label set into two parts: the ground-truth part and the noise part, where the latter is associated with a sparse constraint to make sure the former is clean. Furthermore, we embed the learned unified similarity graph into the process of label disambiguation to restore a more reliable ground-truth label matrix. Finally, the resulting multi-label predictive model is learned with the help of ground-truth label matrix. Extensive experiments on six common used datasets demonstrate that the proposed GMPM achieves comparable performance over the state-of-the-arts.

This work presents an in-situ readout of the transient photoresponse of a graphene-oxide-semiconductor heterostructure by utilizing graphene’s field-effect coupling with the silicon photogate. The reported device which acts as a graphene charge-coupled device (GCCD) pixel is set into pre-avalanche condition by dynamic sinusoidal biasing and then exposed with pulsed illumination. The initial photo-ionized charge packet experiences pre-integration carrier multiplication, boosting the signal-to-noise ratio (SNR) before interacting with the surface traps and defects. The maximum multiplication factor of $\mathrm {\sim }8.5$ and responsivity of $\mathrm {350 A/W}$ are achieved. Arrhenius plots show the viability of operating the device at room temperature as thermal charge contribution is negligible. Moreover, a detailed discussion on reduced power consumption for such a sinusoidal charge-coupled device (CCD) drive concept is also incorporated. The presented technique paves the way for futuristic sinusoidally-driven graphene-silicon-based electron-multiplying CCDs with low-power surveillance and adaptive optics applications.

Engineering vehicle recognition based on video surveillance is one of the key technologies to assist illegal land use monitoring. At present, the engineering vehicle recognition mainly adopts the traditional deep learning model with a large number of floating-point operations. So, it cannot be achieved in edge devices with limited computing power and storage in real time. In addition, some lightweight models have problems with inaccurate bounding box locating, low recognition rate, and unreasonable selection of positive training samples for the small object. To solve the problems, the article proposes an improved lightweight Yolo-Fastest V2 for engineering vehicle recognition fusing location enhancement and adaptive label assignment. The location-enhanced feature pyramid network structure combines deep and shallow feature maps to accurately localize bounding boxes. The grouping k-means clustering strategy and adaptive label assignment algorithm select an appropriate anchor for each object based on its shape and Intersection over Union. The study was conducted on Raspberry Pi 4B 2018 using two datasets and different models. Experiments show that our method achieves the optimal combination in speed and accuracy. Specifically, the mAP50 is increased by 7.02% with the speed of 11.24 fps under the engineering vehicle data obtained by video surveillance in a rural area of China.

Food-related issues have attracted increasing attention recently due to its various applications in our daily life, for example, restaurant service and dietary assessment, which are essential for human health management and understanding food characteristics. Many existing datasets are proposed for various food-relevant tasks, e.g., food detection and segmentation. However, most of the datasets are labelled with only dish-level annotations and lack detailed information for the corresponding ingredients. Hyperspectral imaging (HSI), which can explore the emissive characteristics of different objects with a long region of spectral bands, is of great potential for food and ingredient analysis. In the present work, a new food image dataset HSIFoodIngr-64 containing 3,389 pairs of HSI and RGB images with 21 dish classes and 64 ingredient categories was established. It is the first HSI-based food image dataset, of which all images are labelled with dish-level and ingredient-level annotations and corresponding pixel-wise ingredient masks. Therefore, our dataset can be applied to different food-centric tasks. Furthermore, this research proposed a benchmark method on the food ingredient retrieval task, which consists of two sub-networks called IPSN for ingredient prediction sub-network and ERSN for edge refinement sub-network. Extensive experiments were conducted on HSIFoodIngr-64 dataset and the effectiveness of our proposed method was verified. HSIFoodIngr-64 dataset is expected to provide a new perspective on food analysis and inspire more efforts on various food-related issues. We have made the HSIFoodIngr-64 dataset public at: https://doi.org/10.7910/DVN/E7WDNQ.

The object detection of unmanned aerial vehicle (UAV) images has widespread applications in numerous fields; however, the complex background, diverse scales, and uneven distribution of objects in UAV images make object detection a challenging task. This study proposes a convolution neural network transformer hybrid model to achieve efficient object detection in UAV images, which has three advantages that contribute to improving object detection performance. First, the efficient and effective cross-shaped window (CSWin) transformer can be used as a backbone to obtain image features at different levels, and the obtained features can be input into the feature pyramid network to achieve multiscale representation, which will contribute to multiscale object detection. Second, a hybrid patch embedding module is constructed to extract and utilize low-level information such as the edges and corners of the image. Finally, a slicing-based inference method is constructed to fuse the inference results of the original image and sliced images, which will improve the small object detection accuracy without modifying the original network. Experimental results on public datasets illustrate that the proposed method can improve performance more effectively than several popular and state-of-the-art object detection methods.

Spin-orbit interaction (SOI) of light has attracted enormous interest for its potential applications in controlling light at nanoscale and holds promise for future optical information technology and optoelectronics. In particular, the transverse spin in evanescent waves is regarded as distinct manifestation of SOI and an emerging candidate for miniaturizing optoelectronic circuits. However, further application has been hampered by the inability to conveniently generate light field with configurable transverse spin. Here, we present a simple method to generate free-space light field with arbitrary transverse spin utilizing vector beam interference and numerically demonstrate the configurable spin orbital coupling in nanophotonic waveguide. Chiral waveguide with excellent extinction ratio up to 42 dB is achieved in our simulation. The propagating direction of guided mode can be fully controlled via tuning the ellipticity of transverse spin. The device is compatible with existing integrated nanophotonic devices and may find valuable applications in future optical information processing and optical communication technology.

The automatic vectorization of building shape from very high resolution remote sensing imagery is fundamental in many fields, such as urban management and geodatabase updating. Recently, deep convolutional neural networks (DCNNs) have been successfully used for building edge detection, but the results are raster images rather than vectorized maps and do not meet the requirements of many applications. Although there are some algorithms for converting raster images into vector maps, such vector maps often have too many vector points and irregular shapes. This article proposed a building shape vectorization hierarchy, which combined DCNNs-based building edge detection and a corner extraction algorithm based on principle component analysis for rapidly extracting building corners from the building edges. Experiments on the Jiangbei New Area Buildings and Massachusetts Buildings datasets showed that compared with the state-of-the-art corner detectors, the building vector corners extracted using our proposed algorithm had fewer breakpoints and isolated points, and our building vector boundaries were more complete and regular. In addition, the building shapes extracted using our hierarchy were 7.94% higher than the nonmaximum suppression method in terms of relaxed overall accuracy on the Massachusetts dataset. Overall, our proposed hierarchy is effective for building shape vectorization.

High-density and low-crosstalk waveguide arrays are critical components in optical phased arrays (OPAs) and are widely used in solid-state light detection and ranging (LiDAR). In this work, silicon nitride waveguide superlattices with air gaps are proposed and analyzed theoretically. Mode analysis shows that the introduced air gaps beside the waveguide help shorten the skin depth of the evanescent field and increase the effective index range of the waveguide under single mode conditions. Lower crosstalk is demonstrated between a pair of waveguides with air gaps compared with those without air gaps. On this basis, a waveguide superlattice with air gaps is designed by combining the eigenmode expansion method and particle swarm optimization. The crosstalk of the waveguide superlattice is optimized to −24.3 dB when the waveguide pitch is 0.9 $\mu \text{m}$, the propagation length is 1 mm, and the wavelength is 905 nm. The crosstalk is still below −22.4 dB under the typical process variations and over the wavelength range of 890 nm–920 nm. Therefore, the proposed air-gap waveguide superlattices with high density and low crosstalk offer opportunities to improve the beam steering performance of OPA chips in the near-infrared (NIR) waveband.

Temporal video segmentation and classification have been advanced greatly by public benchmarks in recent years. However, such research still mainly focuses on human actions, failing to describe videos in a holistic view. In addition, previous research tends to pay much attention to visual information yet ignores the multi-modal nature of videos. To fill this gap, we construct the Tencent ‘Ads Video Segmentation’ (TAVS) dataset in the ads domain to escalate multi-modal video analysis to a new level. TAVS describes videos from three independent perspectives as ‘presentation form’, ‘place’, and ‘style’, and contains rich multi-modal information such as video, audio, and text. TAVS is organized hierarchically in semantic aspects for comprehensive temporal video segmentation with three levels of categories for multi-label classification, e.g., ‘place’ - ‘working place’ - ‘office’. Therefore, TAVS is distinguished from previous temporal segmentation datasets due to its multi-modal information, holistic view of categories, and hierarchical granularities. It includes 12,000 videos, 82 classes, 33,900 segments, 121,100 shots, and 168,500 labels. Accompanied with TAVS, we also present a strong multi-modal video segmentation baseline coupled with multi-label class prediction. Extensive experiments are conducted to evaluate our proposed method as well as existing representative methods to reveal key challenges of our dataset TAVS.

In recent years, the deep learning method based on fully convolution networks has proven to be an effective method for the semantic segmentation of remote sensing images (RSIs). However, the rich information and complex content of RSIs make networks training for segmentation more challenging. Specifically, the observing distance between the space-borne cameras and the ground objects is extraordinarily far, resulting in that some smaller objects only occupy a few pixels in the image. However, due to the rapid degeneration of tiny objects during the training process, most algorithms cannot properly handle these common small objects in RSIs with satisfactory results. In this article, we propose a novel feature guide network with a context aggregation pyramid (CAP) for RSIs segmentation to conquer these issues. An innovative edge-guide feature transform module is designed to take advantage of the edge and body information of objects to strengthen edge contours and the internal consistency in homogeneous regions, which can explicitly enhance the representation of tiny objects and relieve the degradation of small objects. Furthermore, we design a CAP pooling strategy to adaptively capture optimal feature characterization that can assemble multiscale features according to the significance of different contexts. Extensive experiments on three large-scale remote sensing datasets demonstrate that our method not only can outperform the state-of-the-art methods for objects of different scales but can also achieve robust segmentation results, especially for tiny objects.

With the rapid development of the social economy, the problem of environmental pollution has been widely concerned. The existing environmental monitoring system adopts a hierarchical centralized management structure, which has some problems, such as data silos and the risk of data falsification. Thus, this paper proposes an environmental monitoring data security model based on the blockchain, which uses the blockchain distributed storage mode to realize the secure sharing of monitoring data and curb the behavior of data forgery. A practical Byzantine fault tolerant mechanism based on credit grouping supervision is adopted to reduce the computation and communication overhead caused by data consensus. Through the cloud chain fusion technology, the encrypted monitoring data is stored on the cloud storage server, and the monitoring data credentials are stored on the blockchain to reduce the storage pressure. And AES(Advanced Encryption Standard) combined with the RSA (Rivest-Shamir-Adleman) encryption algorithm to ensure the security of data transmission and storage. Security analysis and experiments demonstrate that our proposed scheme achieves authenticity, integrity, and security for monitored data. In addition, it effectively reduces the computation, communication, and storage overhead of the block nodes.

The operation characteristics of energy storage can help the distribution network absorb more renewable energy while improving the safety and economy of the power system. Mobile energy storage systems (MESSs) have a broad application market compared with stationary energy storage systems and electric vehicles due to their flexible mobility and good dispatch ability. However, when urban traffic flows rise, the congested traffic environment will prolong the transit time of MESS, which will ultimately affect the operation state of the power networks and the economic benefits of MESS. This paper proposes a bi-level optimization model for the economic operation of MESS in coupled transportation-power networks, considering road congestion and the operation constraints of the power networks. The upper-level model depicts the daily operation scheme of MESS devised by the distribution network operator (DNO) in order to maximize the total revenue of the system. With fuzzy time windows and fuzzy road congestion indexes, the lower-level model optimizes the route for the transit problem of MESS. Therefore, road congestion that affects the transit time of MESS can be fully incorporated in the optimal operation scheme. Both the IEEE 33-bus distribution network and the 29-node transportation network are used to verify and examine the effectiveness of the proposed model. The simulation results demonstrate that the operation scheme of MESS will avoid the congestion period when considering road congestion. Besides, the transit energy consumption and the impact of the traffic environment on the economic benefits of MESS can be reduced.

Landslide detection based on remote sensing images is an effective method for rapidly and accurately detecting landslide regions, which can aid in disaster prevention and mitigation. Landslide detection methods based on semantic segmentation can be used to delineate the scope of landslides while detecting their location. Most existing models use multitemporal or geological data to improve accuracy. However, the large amount of data introduces additional parameters, consuming significant computing resources. Therefore, this article proposes Reg-SA–UNet++, a model for landslide detection, which uses a single-temporal image captured postlandslide. Reg-SA–UNet++ is based on UNet++ with the following modifications: deep supervised pruning is removed for fewer model parameters and increased detection accuracy; RegNet is employed to replace the convolutional blocks in the encoding process to reduce the number of parameters and improve feature acquisition and attention modules are added at the connection of the convolutional blocks of each layer to strengthen the model's attention to landslide features. The overall accuracy and F1 score of the Reg-SA–UNet++ model for the constructed landslide dataset (93.37% and 92.41%, respectively) and landslide mapping (97.09% and 96.10%, respectively) verify the effectiveness of the proposed model in detecting landslides from remote sensing images.

The uncertainty of user-side resource response will affect the response quality and economic benefit of load aggregator (LA). Therefore, this paper regards the flexible user-side resources as a virtual energy storage (VES), and uses the traditional narrow sense energy storage (NSES) to alleviate the uncertainty of VES. In order to further enhance the competitive advantage of LA in electricity market transactions, the opertion mechanism of LA in day-ahead and real-time market is analyzed, respectively. Besides, truncated normal distribution is used to simulate the response accuracy of VES, and the response model of NSES is constructed at the same time. Then, the hierarchical market access index (HMAI) is introduced to quantify the risk of LA being eliminated in the market competition. Finally, combined with the priority response strategy of VES and HMAI, the capacity allocation model of NSES is established. As the capacity model is nonlinear, Monte Carlo simulation and adaptive particle swarm optimization algorithm are used to solve it. In order to verify the effectiveness of the model, the data from PJM market in the United States is used for testing. Simulation results show that the model established can provide the effective NSES capacity allocation strategy for LA to compensate the uncertainty of VES response, and the economic benefit of LA can be increased by 52.2% at its maximum. Through the reasonable NSES capacity allocation, LA is encouraged to improve its own resource level, thus forming a virtuous circle of market competition.

Remote sensing (RS) image change detection analyzes and determines the difference of surface states at different times. Polygons are common vector data to express the surface states in geographic information systems. However, the production and updating of polygons typically depend on manual procedures, which costs significant time and human resources. This article proposes an automatic polygon change detection method based on high-resolution RS images and deep learning image classification. First, bitemporal images are segmented by boundary-preserved masking simple linear iterative clustering based on the boundaries and categories of polygons. Then, the labeled dataset is automatically generated by adaptively cropping image segments according to the center-to-boundary distance. Date-1 data are used to train the model and purified by two different classifiers to improve the model performance. Date-2 data, whose categories are predicted, are the basic units of change detection analysis. Furthermore, an improved bilinear convolutional neural network with a reduced number of parameters and training time is applied to fine-grained classification. Finally, the changed polygons can be retrieved based on category comparison and post-processing. In our experiments, two real datasets with high-resolution RS images and land-use polygons are employed to validate the effectiveness of our proposed method. The results are examined by visual interpretation and show that the precision and recall rates for dataset 1 are 90.5% and 93.1%, respectively, while those for dataset 2 are 87.6% and 90.8%. Through our method, the existing polygons can be used appropriately, and the difficulty of data updating can be reduced.

A thermal hydraulic disaggregation of soil moisture (THySM) algorithm was implemented to downscale NASA's soil moisture active passive (SMAP) enhanced soil moisture (SM) product to 1 km over the continental United States (CONUS). This algorithm was developed by combining thermal inertia theory with a soil hydraulic-based approach that considers fine-scale SM spatial distribution driven by both heat fluxes and hydraulic conductivity in soils. Relative soil wetness values were estimated using land surface temperature and normalized difference vegetation index for the thermal inertia model and using soil properties for the hydraulic model. The relative soil wetness values at 1 km from both models were then combined by using weighting functions whereby the spatial distribution of SM was governed more by thermal fluxes during times of strong heat transport and infiltration during moisture abundant soil conditions. THySM values were evaluated using in situ SM measurements from SMAP Core Validation Sites (CVS), the US Department of Agriculture Soil Climate Analysis Network, and the National Oceanic and Atmospheric Administration Climate Reference Network over CONUS. THySM shows higher accuracy than the SMAP / Sentinel-1 (SPL2SMAP_S) 1 km SM product when compared to in situ measurements. The accuracy of THySM is 0.048 m3/m3 based on unbiased root mean square error (ubRMSE), outperforming SPL2SMAP_S by 0.01–0.02 m3/m3. The ubRMSE of THySM 1 km SM over the SMAP grassland/rangeland-dominated CVS sites is better than 0.04 m3/m3, which meets the SMAP mission SM accuracy requirement applied at 9 and 36 km.

A PCB spring terminal is a common type of connector found on the printed circuit board of electronic devices. Contact force is a crucial performance metric of this component that directly affects the product’s dependability. In actual production, the contact force is unstable, and the influencing factors are unknown, whereas conventional experimental research is restricted to costly and error-prone trial-and-error procedures. In this study, finite element simulation technology and Spearman rank correlation analysis are combined to determine the crucial factors controlling the contact force of the PCB spring terminals in the structural design parameters and the plating related parameters. The validity of the results is confirmed by the measured data. The results indicate that the thickness of the spring terminal is the most significant factor in controlling the contact force. According to the conclusion, the company has re-established the inspection standard and resolved production problems. This case shows that the combination of finite element simulation and correlation analysis can be used to find the most important design parameters of such components quickly, accurately, and cheaply.

The signal-to-noise ratio (SNR) is important observations in global navigation satellite system-reflectometry (GNSS-R) technology. The oscillation frequency in the SNR arc is sensitive to different reflecting surfaces and can be used to build height model to track the variation of snow depth. However, it is difficult to obtain retrieval results with snow depth of zero in the actual snow depth retrieval experiments based on GNSS-R technology, which indicates that the classical model has nonnegligible retrieval errors in the snow-free state. This study aims to realize the detection of ground truth information before snow depth retrieval, i.e., classification of snow-free state and snow-covered state. Machine learning was introduced to achieve the aforementioned purpose and the SNR arc was used as the input data. Compared with the current mainstream topography correction algorithms, the algorithm proposed in this study does not rely on any priori ground measured data and has theoretical universality. The detection results can constrain the retrieval snow depth in the snow-free state and, thus, improve the retrieval accuracy. The experimental results for the 2014 seasonal snowpack at P351 station in Idaho, USA, show that the detection results obtained based on support vector machines agree well with the measured snow depth provided by the SNOTEL network, and the overall detection accuracy can reach about 96%. The daily snowpack state is determined by the majority of SNR arcs detected during the day and is only considered reliable if the percentage exceeds 75%. Only one day of the detection results was inaccurate and only 8 days (8/365) did not reach the set threshold of 75%. With the help of the detection results, the root-mean-square error of snow depth retrieval can be reduced from 20 cm in the classical algorithm to 15 cm, which results in a 25% improvement in retrieval accuracy. Moreover, this study broadens the application value of GNSS signals and provides a reference for the application of SNR in the detection field.

The routing congestion over a QC-LDPC decoder with a large circular permutation matrix (CPM) size has long been an obstacle to high throughput designs. This paper presents a large-CPM congestion-free decoder for (18396, 16416) quasi-cyclic Euclidean geometry low-density parity-check (QC-EG-LDPC) code in NAND flash application. Considering area efficiency in scheduling schemes and the array dispersion structure, the Array-Disperse Based Dual Variable Node Unit (VNU) Cluster Architecture fully leverages the code structure to support at least two physical channels of the Open NAND Flash Interface 5.0 (ONFI 5.0). In addition, the proposed congestion-aware analysis and implementation method achieve a highly parallel decoder at a 70% utilization ratio. Implemented in TSMC 28nm process, the presented decoder provides 38.64 Gbps throughput at RBER=1.456% Bi-AWGN channel with an area of 2.97 mm2.

Owing to its advantages of low-cost and fast detection, pulsed thermography has become a promising technique to detect subsurface defects in materials of carbon fiber reinforced polymer (CFRP). Since defect signals in the detected results always suffer from low contrast due to the instability of detecting environment, feature extraction methods are required to enhance the visualization of defects. However, many state-of-the-art feature extraction methods have difficulties in overcoming the interference from noise and background, so that their effects are limited in highlighting the defects. To solve this problem, a novel methodology of combining signal filtering with feature extraction is proposed in this paper. In this approach, thermal images are first smoothed by a difference of Gaussian convolutional (DoGC) filters, which is designed to eliminate noise and uneven background based on their frequencies. Furthermore, the method of sparse principal component thermography (SPCT) is adopted to extract the features of defects. Two experiments on sample laminates have suggested that, DoGC-SPCT is superior to other feature extraction methods in the following aspects. Firstly, the DoGC filter can effectively eliminate most of the interference, thus facilitating defect identification during the process of feature extraction. Secondly, the computational outcomes show that DoGC-SPCT leads to higher values in the index of signal to noise ratios for the defects. Finally, DoGC-SPCT leads to higher interpretability, which has smoother background in the obtained results.

Unmanned aerial vehicles (UAVs) equipped with lightweight sensors, such as RGB cameras and LiDAR, have significant potential in precision agriculture, including object detection. Tassel detection in maize is an essential trait given its relevance as the beginning of the reproductive stage of growth and development of the plants. However, compared with general object detection, tassel detection based on RGB imagery acquired by UAVs is more challenging due to the small size, time-dependent variable shape, and complexity of the objects of interest. A novel algorithm referred to as YOLOv5-tassel is proposed to detect tassels in UAV-based RGB imagery. A bidirectional feature pyramid network is adopted for the path-aggregation neck to effectively fuse cross-scale features. The robust attention module of SimAM is introduced to extract the features of interest before each detection head. An additional detection head is also introduced to improve small-size tassel detection based on the original YOLOv5. Annotation is performed with guidance from center points derived from CenterNet to improve the selection of the bounding boxes for tassels. Finally, to address the issue of limited reference data, transfer learning based on the VisDrone dataset is adopted. Testing results for our proposed YOLOv5-tassel method achieved the mAP value of 44.7%, which is better than well-known object detection approaches, such as FCOS, RetinaNet, and YOLOv5.

An error floor phenomenon, decoding performance, and throughput are three major concerns for LDPC decoders in NAND Flash applications. With a penalty method and an active iteration mechanism, we present a Unified Penalty Gradient Descent Bit Flipping (UP-GDBF) decoding algorithm, which not only possesses error-floor free property but also improves convergence speed in decoding performance. To fulfill the high-throughput requirement while maintaining reliable error correction capability, we propose an energy-based backtracking scheme to reduce 40% latency with a negligible 0.8% area overhead. Implemented in TSMC 16nm process, the proposed 4KB LDPC decoder can achieve a throughput of 19.3 Gbps with 0.120 mm2 area to satisfy ONFI 5.0 throughput requirement. Compared to existing approaches, our decoder architecture provides superior data rate and decoding performance in both 1KB and 4KB LDPC codes.

A novel voltage control scheme for an interior permanent magnet synchronous motor (IPMSM) servo drive in the constant power region with intelligent estimation of the motor parameter is proposed in this study. In the novel voltage control scheme, a feedforward voltage angle controller is proposed where an intelligent parameter estimation method by using a wavelet fuzzy neural network (WFNN) is developed to estimate the $q$ -axis inductance online. In this study, in order to minimize the copper loss, a flux-weakening (FW) control scheme under maximum phase voltage is developed first. Then, an adaptive backstepping based nonlinear controller (ABNC) considering nonzero $d$ -axis current is developed to improve the robustness of the speed control. The Lyapunov stability theorem is used to derive the adaptive law of the online estimation of the lumped uncertainty to ensure the asymptotical stability of the ABNC. Moreover, a feedforward voltage angle controller is developed for the voltage control where the $q$ -axis current controller is retained in order to ensure the steady-state performance of the control system. Furthermore, the WFNN is adopted to estimate the actual $q$ -axis inductance value online for the feedforward voltage angle controller to improve the dynamic response. In addition, some experimental results are demonstrated to verify the effectiveness of the proposed voltage control scheme with ABNC in the constant power region.

A distributed secondary control (DSC) strategy that combines $Q$-learning and pinning control is originally proposed to achieve a fully optimal DSC for droop-controlled microgrids (MGs). It takes advantages of cross-fusion of the two algorithms to realize the high efficiency and self-adaptive control in MGs. It has the following advantages. Firstly, it adopts the advantages of reinforcement learning in autonomous learning control and intelligent decision-making, driving the action value of pinning control for feedback adaptive correction. Secondly, only a small part of points selected as pinned points needs to be controlled and pre-learned, hence the actual control problem is transformed into a synchronous tracking problem and the installation number of controllers is further reduced. Thirdly, the pinning matrix can be modified to adapt to plug-and-play operation under the distributed control architecture. Finally, the effectiveness and versatility of the proposed strategy are demonstrated with a typical droop-controlled MG model.