Harman (China)

This paper presents a rear obstacle detection system by using a single rear view camera. The system can detect various static and moving obstacles behind the cars. An efficient hierarchical detecting strategy is used to achieve high detection rate and low false positives. The temporal Inverse perspective mapping difference image based coarse detection is used to estimate whether there are obstacles in the predetermined warning area at first stage. Then a novel integral image based segmentation algorithm is developed for fine obstacle segmentation. Finally, the blob analysis is utilized for obstacle representation and verification. Our system achieves 94.2% detection rate and 16% false positives rate on 125 challenging video sequences. The average processing speed of the system is 25fps on a standard laptop.

As protective shells for their biological functions against environmental damage and attack by natural predators, the silkworm (Bombyx mori) cocoon and its wild partner have distinctive multi-layer structures, which are systematically studied in this paper by the SEM, the thermogravimetric analyzer, and the Fourier transform infrared spectroscopy. Their mechanical properties are also investigated for the whole hierarchy and each cascade as well. In order to better demonstrate the superior survivability of cocoons in harsh environments, air permeability and moisture vapor transmission rate of the silkworm cocoon are tested. The results show the silkworm cocoons have excellent air permeability and moisture vapor transmission rate. A better understanding of different cocoons? bio-functions will be of particular importance to design thermal textiles and provide better comfort and safety for clothing in future.

This paper present an efficient real time rectangle speed limit sign recognition system. The system design considers computation load and hardware resources for driver assistant system. First multi-scale overlapping LBP features are used to train AdaBoost cascade classifier for speed limit sign object detection. Then a simple linear prediction method is used to do tracking task. At the recognition stage, a novel efficient algorithm is used to correct rotation angle, and then integral image based adaptive threshold algorithm is adopted to segment the speed limit number. The clustering based binary tree of linear support vector machine is adopted for classification task. The system is tested on real road scene video sequences. It achieves 98.3% recognition rate with approximate 16 fps frame rate on laptop.

Many brain regions exhibit rhythmical activity thought to reflect the summed behaviour of large populations of neurons. The endogenous alpha rhythm has been associated with phase-dependent modulation of corticospinal excitability. However, whether exogenous alpha rhythm, induced using transcranial alternating current stimulation (tACS) also has a phase-dependent effect on corticospinal excitability remains unknown. Here, we triggered transcranial magnetic stimuli (TMS) on the up- or down-going phase of a tACS-imposed alpha oscillation and measured motor evoked potential (MEP) amplitude and short-interval intracortical inhibition (SICI). There was no significant difference in MEP amplitude or SICI when TMS was triggered on the up- or down-going phase of the tACS-imposed alpha oscillation. The current study provides no evidence of differences in corticospinal excitability or GABAergic inhibition when targeting the up-going (peak) and down-going (trough) phase of the tACS-imposed oscillation.

Voice Activity Detection (VAD) plays an important role in audio processing, but it is also a common challenge when a voice signal is corrupted with strong and transient noise. In this paper, an accurate and causal VAD module using a long short-term memory (LSTM) deep neural network is proposed. A set of features including Gammatone cepstral coefficients (GTCC) and selected spectral features are used. The low complex structure allows it can be easily implemented in speech processing algorithms and applications. With carefully pre-processing and labeling the collected training data in the classes of speech or non-speech and training on the LSTM net, experiments show the proposed VAD is able to distinguish speech from different types of noisy background effectively. Its robustness against changes including varying frame length, moving speech sources and speaking in different languages, are further investigated.

When the incident signals received by a phased array are correlated, spatial smoothing techniques which are used in element space (ES) to overcome the rank deficiency of the signal subspace are not able to be directly applied in beam space (BS). Recently it has been shown that an approximate ES covariance matrix with the appropriate signal and noise subspaces can be reconstructed from the BS covariance matrix thus allowing spatial smoothing to be carried out on the reconstructed ES covariance matrix and hence high resolution BS direction‐of‐arrival estimates to be obtained for correlated arrivals. The theory previously developed required that the beams be orthogonal and this study shows how the use of the Moore–Penrose pseudo‐inverse and sector focused stability version of BS processing allows the requirement of orthogonality to be reduced to the more realistic and practical constraint that the beams used be independent. The special case of a uniform linear array is considered and it is shown that due to the Vandermonde structure of the beamforming matrix spatial smoothing can be carried out directly in BS.

Apart from MIMO, high resolution techniques for phased array radars such as optimum beamforming and MUSIC ignore the transmit beam pattern and work solely with the receiver outputs. However for directional transmit phased arrays which scan a relatively narrow beam over a restricted sector of interest ignoring the transmit beam pattern leads to model mismatch. The consequent losses can be mitigated by stacking the receiver outputs as the transmit beam is scanned over the sector of interest and modifying the steering vectors for the subsequent processing. This proposed hybrid approach can be interpreted as a MIMO virtual array system achieving higher beam output SNRs whilst still providing the MIMO ability to resolve more DOAs than conventional phased arrays. Numerical and simulation examples are presented to illustrate the stacked approach and to compare the different approaches to estimating the DOAs of uncorrelated signals.

Since the founding of China in 1949, policies in foreign languages have undergone earth-shaking changes, as well as continuous innovation and development. Until the Cultural Revolution happened in 1966, foreign language education suffered a lot of trauma, and foreign language education policy also stopped. Based on the previous studies of foreign language education policies in this period, this paper expounds on the development and changes in foreign language education policies in the past 16 years through the historical context. It found that people also play an enlightening role in the formulation of foreign language education policies in terms of politics, compulsory education and the international environment. The change and improvement of foreign language education policy during this period can be divided into two stages: the period of 1949-1957 when Russian was the main language, supplemented by English, and the period of 1958-1965 when English was the main language, and the period of Russian development. People can have a clearer understanding of the changes in education in foreign language policies.

Cluster tools play a significant role in the entire process of wafer fabrication. Wafer residency time constraints and chamber cleaning requirements are commonly seen in etching, chemical vapor deposition, coating processes, etc. They make the scheduling problem of cluster tools more challenging. This work aims to provide a solution for dual-arm cluster tools with wafer residency time constraints and chamber cleaning requirements. To do so, it proposes a novel virtual wafer-based scheduling method. By this method, under a steady state, a process module (PM) processes either a real or virtual wafer at a time. When a PM processes a virtual one, its chamber performs a cleaning operation. In this way, we can meet not only the strict residency time constraints for real wafers, but also innovatively performs chamber cleaning operations as required. Based on such a novel scheduling method, an efficient binary integer programming model is established to maximize the throughput of cluster tools. Finally, experiments are performed to show the efficiency and effectiveness of the proposed method.

In mountain tunnel blasting construction, challenges such as over-excavation and improper particle size distribution are frequently encountered. Traditional neural network prediction models and empirical formulas have proven inadequate for optimizing construction parameters. To improve the accuracy of prediction models, this study employed Principal Component Analysis (PCA) to identify three key factors influencing construction outcomes as input variables for the model. Additionally, Particle Swarm Optimization (PSO) was integrated for parameter adjustment, leading to the optimization of the parameters within the Deep Belief Network (DBN) model. Two PCA-PSO-DBN models were developed, specifically addressing tunnel over-excavation and the equivalent size of crushed rocks. By training and predicting data from Sect. 3 of the NEOM New City tunnel project, the feasibility of the model was validated through on-site data analysis. The results indicated that compared to traditional DBN and PCA-DBN models, the proposed model reduced maximum errors by 16.98, 7.68, and 11.37, 4.85%, respectively, demonstrating higher precision. Following blasting parameter optimization, the reductions in maximum linear over- or under-excavation and the equivalent size of crushed rocks in the tunnel reached 40.94% and 18.70%, respectively, compared to the original blasting plan. This model introduces innovative methodologies and possibilities, offering valuable insights and references for similar endeavors.

Speed limit sign recognition is one of the important components for a driver assistance system.An efficient real time rectangular speed limit sign recognition system was proposed.The system framework design considered the computation load and hardware resources for a driver assistance system.First,multi-scale overlapping local binary pattern(LBP) image features were used to train an AdaBoost cascade classifier for sign detection.Then a simple linear prediction method was used to do the tracking task.In the recognition stage,the projection method was used to correct the rotation angle and then the integral image based adaptive threshold algorithm was applied to segment the speed limit number,and then the principal component analysis(PCA) was used for feature vector extraction.Finally,a clustering based binary tree of a linear support vector machine was designed for the classification task.The system achieved a 98.3% recognition rate with an approximate frame rate of 16fps in video files for the laptop computer system during actual road tests.

In this paper, an Extended Kalman Filter (EKF) - based visual SLAM algorithm using single fisheye lens camera to build a large scale sense is presented. The primary contribution of this work is the adaption of MonoSLAM from conventional perspective cameras and wide angle cameras to fisheye lens cameras in which the Polynomial Camera Model [10] is adopted to obtain 3D information. For data association, pSIFT [1] is implemented, which is designed dedicatedly for feature matching among fisheye images. In the experiment, we implement sub-mapping algorithm [5] and optimization jointly over all the frames with 3D feature position and camera poses estimated by MonoSLAM. The result confirms that our algorithm can work with fisheye camera well.

Abstract Detrital composition of sandstone is the most important data for siliciclastic studies including sandstone classification, provenance analysis, oil and gas exploration. A large amount of detrital composition data has accumulated over the past decades, however, they are scattered in publications without unified standards. Here we constructed a global dataset of detrital components of sandstones from 646 peer‐reviewed publications using Gazzi‐Dickinson method. A total of 19,861 samples from Precambrian to Quaternary are involved in this dataset. For each sample, we present details on reference information, geographic information, geological background, depositional age and the original data. It is a high‐quality dataset for the information on each sandstone sample from different studies which was standardized. The dataset can be used widely, such as for stratigraphic comparison, provenance analysis, exploring the general laws of the source‐to‐sink process and geological engineering.

In today’s world, many fields are supported with AI, but there are few research about the applications of large language models on sentencing crime. This paper formulates a legal-large language model specialized in Chinese criminal adjudication. The objective of the model is to give charge recommendations and sentencing range that is in accordance to China’s laws. It is based on a transformer-based model (Qwen-7B) and fine-tuned on more than 500,000 anonymized judicial documents. The data we used are from CAIL2018 and CAIL2020, two large legal datasets that store huge amounts of legal cases. Our model addresses charge prediction and sentencing recommendation tasks through the incorporation of prompt-based instruction tuning to simulate judicial logic. The model is able to receive long case inputs and give meaningful outputs. It shows high precision in determining charges and suggesting sentencing ranges. The model also shows high capability in aligning outputs with laws and statutes, giving the charges and sentencing ranges according to legal statutes. Interpretive pattern analysis of the model indicates capabilities in statutory fit. The model shows weakness in addressing complicated case details, where the model will show decrease in precision of output. The results of our model provide useful conclusions for the development of AI-supported judicial systems, where Large Language Models can help and support judges’ decision.

Intelligent fault diagnosis of train gearbox plays a crucial role in intelligent operation and maintenance system of a train. Traditional deep learning diagnostic models are generally limited to training with known fault samples. However, in real-world scenarios, unknown new faults often occur in the application of the trained diagnostic models. Conventional diagnostic models often fail to correctly identify the category of unknown faults, mistakenly classifying them as known categories, which brings significant potential risks. To address the open-set recognition task in train gearbox fault diagnosis, this paper proposes a diagnostic model based on generative flow (Glow) model and generative adversarial networks (GAN), which is termed GlowGAN. Specifically, the proposed model innovatively adopts a modular design, skillfully combining the Glow model with the GAN. The high-quality images generated by the Glow are used to train a stable and performant discriminator of the GAN. Subsequently, a residual network (ResNet) is employed to extract deep features from the images, which are then passed to the discriminator to obtain sample confidence score. Samples with low confidence score are classified as unknown faults. The proposed method is verified on two train gearbox datasets that the proposed GlowGAN can accurately identify unknown faults and outperforms existing open-set recognition methods.

In this study, a more efficient time domain (TD) virtual rotating array (VRA) method is proposed that employs a zeroth-order interpolation scheme. The interpolation of sound fields in the receiver plane using only the nearest microphone at each time step improves the efficiency of the entire process. Additionally, the VRA method is demonstrated to function well with near-field acoustic holography for localizing rotating sound sources. The proposed methods are numerically and experimentally validated. The zeroth-order interpolation scheme is compared with linear, barycentric, and radial basis function schemes using either a circular array or an arbitrary array. The holographic reconstructions are compared for selected one-third octave band center frequencies. Compressive sensing-based holography is used for increasing the resolution of the numerical simulations at low frequencies. The source strength is calculated by integrating the holographic spectra in the source plane. A comparison of the predicted source strengths and locations suggests that the zeroth-order scheme yields a more accurate solution than higher-order schemes. The proposed zeroth-order scheme and the source reconstruction using VRA signals have the potential to visualize sound fields produced by compact rotating structures.

The health condition of rolling bearings is crucial for the safety of rotating machines in industry. However, the bearing abnormal samples are rare in engineering practice, thus the historical bearing dataset is usually imbalanced between the normal and abnormal samples, which degrades the performance of traditional intelligent diagnosis model. This paper proposes an interpretable time-frequency spectra augmentation and enhancement (TFSAE) method for bearing imbalanced fault diagnosis. The TFSAE explores a varying-parameter wavelet transform technique to expand the amounts of time-frequency spectra of bearing abnormal samples, and designs a time-frequency attention mechanism to enhance the time-frequency features. The data augmentation technique is physically interpretable, and the enhanced features have physical meanings. Specifically, each sample of the abnormal data is firstly converted to several time-frequency spectra by continuous wavelet transform with varying wavelet bandwidths. Then, the time-frequency spectra of the abnormal samples are combined with those of normal samples using a fixed-parameter wavelet transform to construct balanced bearing dataset, which is used to train a fault diagnosis model to achieve unbiased fault diagnosis. Finally, the time-frequency attention mechanism is designed in the fault diagnosis model to enhance the time-frequency features relevant to fault characteristics while alleviating noise contamination. Experimental results on two bearing datasets demonstrate that the proposed method achieves significant improvements over the state-of-the-art methods in bearing imbalanced fault diagnosis.

AVB is a real time protocol and now a days it is preferred in Automotive world in Infotainment ECUs for Audio communication over Ethernet as it insures low latency and real time aspects while Audio data transmission over Ethernet. Media clock recovery is a mechanism for achieving a glitch-free audio experience over Ethernet. A complete software-based MCR mechanism requires a lot of Software tuning and is prone to system latencies, whereas a hardware-based MCR requires costly electronic components. The proposed Media clock recovery mechanism in this study combines both hardware and software-based media clock recovery (MCR) mechanisms. The hardware part consists of a clock multiplier, which can be considered as a PLL but lacks phase alignment, effectively making it a frequency syntonization. Therefore, using software components, this phase alignment is achieved, making it a complete frequency synchronization. Keywords— Automotive AVB, Ethernet, MCR, Hardware-Software Co-design

THE great channels of trade in North America run east and west. The great river systems of the continent run north and south. There is, however, one striking exception to this general rule, where the course of the Great Lakes and the St. Lawrence breaks through the Appalachian Range, and forms a continuous waterway, 2000 miles in length, from the center of the continent to the Atlantic Ocean. Much of this water course is now open to navigation and the American Great Lakes have within the last twenty years witnessed the most remarkable maritime developments of any section in the world. The Lakes extend approximately 1000 miles from Duluth or Chicago to Buffalo through the very heart of America; and within the last two decades there has grown up on these Lakes a traffic whose tonnage exceeds that of the Mediterranean and the Black Sea combined; indeed the movement of vessels through the locks between Lake Superior and Lake Huron is twice the combined movement of vessels through the Suez and Panama Canals, and more tonnage passes Detroit in nine months than clears from New York or Liverpool in a year. Along or near these Great Lakes lives approximately forty per cent of the population of the United States. Not only are the shores of the Lakes thickly populated, but the territory contiguous to them is rich in agriculture and in mineral products. Wheat, grain, livestock, iron, coal and copper are among the great inheritance of this rich fertile region of our country. This region has also become a great manufacturing center. Flour, foodstuffs, packing products, automobiles, rails and other heavy steel products, and many other articles of commerce are produced in this region; and these articles, as well as the products of the soil and the mines, flow eastward over the waters of the Great Lakes until the port of Buffalo is reached, where they must be transferred to the rails, and move the last 500 miles of their journey to the seaboard by car rather than by boat.

The Hanjialing pluton occurs within a Paleoproterozoic rift belt in the Liaodong Peninsula in Northeast China, a region endowed with rich polymetallic mineral deposits including gold, lead, and zinc. The genesis and distribution of these ores are closely related to the formation of the pluton. Here, we reconstruct the three‐dimensional geological characteristics of the Hanjialing pluton in an attempt to constrain the relationship between intrusion and mineralization. Our high‐precision gravity and magnetic measurements and magnetotelluric profiles trace the three‐dimensional geometry of the pluton as an inflated balloon. We also demarcate its lateral and vertical extension and characterize the distribution of the associated ore mineralization. Our data suggest that more than 8 km of the upper part of the intrusion has been removed by erosion since the Cretaceous.