Cold Start Problem of Vehicle Model Recognition under Cross-Scenario Based on Transfer Learning

As a major function of smart transportation in smart cities,vehicle model recognition plays an important role in intelligent transportation.Due to the difference among different vehicle models recognition datasets,the accuracy of network model training in one scene will be greatly reduced in another one.However,if you don’t have a lot of vehicle model datasets for the current scene,you cannot properly train a model.To address this problem,we study the problem of cold start of vehicle model recognition under cross-scenario.Under the condition of small amount of datasets,combined with the method of transfer learning,load the DAN(Deep Adaptation Networks)and JAN(Joint Adaptation Networks)domain adaptation modules into the convolutional neural network AlexNet and ResNet,and get four models:AlexNet-JAN,AlexNet-DAN,ResNet-JAN,and ResNet-DAN which can achieve a higher accuracy at the beginning.Through experiments,transfer the vehicle model recognition from the network image dataset(source domain)to the surveillance-nature dataset(target domain),both Top-1 and Top-5 accuracy have been improved by at least 20%.

Sika Deer Facial Recognition Model Based on SE-ResNet

The scale of deer breeding has gradually increased in recent years and better information management is necessary,which requires the identification of individual deer.In this paper,a deer face dataset is produced using face images obtained from different angles,and an improved residual neural network(ResNet)-based recognition model is proposed to extract the features of deer faces,which have high similarity.The model is based on ResNet-50,which reduces the depth of the model,and the network depth is only 29 layers;the model connects Squeeze-and-Excitation(SE)modules at each of the four layers where the channel changes to improve the quality of features by compressing the feature information extracted through the entire layer.A maximum pooling layer is used in the ResBlock shortcut connection to reduce the information loss caused by messages passing through the ResBlock.The Rectified Linear Unit(ReLU)activation function in the network is replaced by the Exponential Linear Unit(ELU)activation function to reduce information loss during forward propagation of the network.The preprocessed 6864 sika deer face dataset was used to train the recognition model based on SEResnet,which is demonstrated to identify individuals accurately.By setting up comparative experiments under different structures,the model reduces the amount of parameters,ensures the accuracy of the model,and improves the calculation speed of the model.Using the improved method in this paper to compare with the classical model and facial recognition models of different animals,the results show that the recognition effect of this research method is the best,with an average recognition accuracy of 97.48%.The sika deer face recognition model proposed in this study is effective.The results contribute to the practical application of animal facial recognition technology in the breeding of sika deer and other animals with few distinct facial features.

Construction of multi-factor identification model for real-time monitoring and early warning of mine water inrush

As a new technical means that can detect abnormal signs of water inrush in advance and give an early warning,the automatic monitoring and early warning of water inrush in mines has been widely valued in recent years.Due to the many factors affecting water inrush and the complicated water inrush mechanism,many factors close to water inrush may have precursory abnormal changes.At present,the existing monitoring and early warning system mainly uses a few monitoring indicators such as groundwater level,water influx,and temperature,and performs water inrush early warning through the abnormal change of a single factor.However,there are relatively few multi-factor comprehensive early warning identification models.Based on the analysis of the abnormal changes of precursor factors in multiple water inrush cases,11 measurable and effective indicators including groundwater flow field,hydrochemical field and temperature field are proposed.Finally,taking Hengyuan coal mine as an example,6 indicators with long-term monitoring data sequences were selected to establish a single-index hierarchical early-warning recognition model,a multi-factor linear recognition model,and a comprehensive intelligent early-warning recognition model.The results show that the correct rate of early warning can reach 95.2%.

Online hierarchical recognition method for target tactical intention in beyond-visual-range air combat

Online accurate recognition of target tactical intention in beyond-visual-range (BVR) air combat is an important basis for deep situational awareness and autonomous air combat decision-making, which can create pre-emptive tactical opportunities for the fighter to gain air superiority. The existing methods to solve this problem have some defects such as dependence on empirical knowledge, difficulty in interpreting the recognition results, and inability to meet the requirements of actual air combat. So an online hierarchical recognition method for target tactical intention in BVR air combat based on cascaded support vector machine (CSVM) is proposed in this study. Through the mechanism analysis of BVR air combat, the instantaneous and cumulative feature information of target trajectory and relative situation information are introduced successively using online automatic decomposition of target trajectory and hierarchical progression. Then the hierarchical recognition model from target maneuver element, tactical maneuver to tactical intention is constructed. The CSVM algorithm is designed for solving this model, and the computational complexity is decomposed by the cascaded structure to overcome the problems of convergence and timeliness when the dimensions and number of training samples are large. Meanwhile, the recognition result of each layer can be used to support the composition analysis and interpretation of target tactical intention. The simulation results show that the proposed method can effectively realize multi-dimensional online accurate recognition of target tactical intention in BVR air combat.

REVERSE MODELING FOR CONIC BLENDING FEATURE

A novel method to extract conic blending feature in reverse engineering is presented. Different from the methods to recover constant and variable radius blends from unorganized points, it contains not only novel segmentation and feature recognition techniques, but also bias corrected technique to capture more reliable distribution of feature parameters along the spine curve. The segmentation depending on point classification separates the points in the conic blend region from the input point cloud. The available feature parameters of the cross-sectional curves are extracted with the processes of slicing point clouds with planes, conic curve fitting, and parameters estimation and compensation, The extracted parameters and its distribution laws are refined according to statistic theory such as regression analysis and hypothesis test. The proposed method can accurately capture the original design intentions and conveniently guide the reverse modeling process. Application examples are presented to verify the high precision and stability of the proposed method.

Method to generate training samples for neural network used in target recognition

Training neural network to recognize targets needs a lot of samples.People usually get these samples in a non-systematic way,which can miss or overemphasize some target information.To improve this situation,a new method based on virtual model and invariant moments was proposed to generate training samples.The method was composed of the following steps：use computer and simulation software to build target object＇s virtual model and then simulate the environment,light condition,camera parameter,etc.;rotate the model by spin and nutation of inclination to get the image sequence by virtual camera;preprocess each image and transfer them into binary image;calculate the invariant moments for each image and get a vectors＇ sequence.The vectors＇ sequence which was proved to be complete became the training samples together with the target outputs.The simulated results showed that the proposed method could be used to recognize the real targets and improve the accuracy of target recognition effectively when the sampling interval was short enough and the circumstance simulation was close enough.

几种开源英语识别工具包的对比分析

对开源英语语音识别工具包在可用性和识别准确性方面进行了对比评价。所对比的语音工具包为HTK语音工具包、CMU Sphinx系列语音处理系统和Kaldi语音工具包。通过对比分析发现,Kaldi语音工具包提供了最先进和全面的声学模型训练技术支持,具有最出色的识别正确率结果;CMU Sphinx系列语音工具包在提供较为全面的声学模型训练基础上,具有最好的识别效率;而HTK语音识别工具包所提供的训练技术支持最少,并且需要自行开发训练脚本,因此使用难度最大。

建立河流水质模型的状态空间分析法

该文导出了河流水质系统的离散时间状态空间模型和随机状态模型，采用最小二乘法和三种定阶方法，对水质模型中分布延迟系数和阶次进行估计。通过计算机仿真计算和建模实例，证实了辨识方法的正确性和状态空间模型的可靠性。

Rapid determination of tannin in Danshen and Guanxinning injections using UV spectrophotometry for quality control

A technique for the determination of tannin content in traditional Chinese medicine injections(TCMI)was developed based on ultraviolet(UV)spectroscopy.Chemometrics were used to construct a mathematical model of absorption spectrum and tannin reference content of Danshen and Guanxinning injections,and the model was veried and applied.The results showed that the established UV-based spectral partial least squares regression(PLS)tannin content model performed well with a correlation coefficient(r)of 0.952,root mean square error of calibration(RMSEC)of 0.476
g/ml,root mean square error of validation(RMSEV)of 1.171
g/ml,and root mean square error of prediction(RMSEP)of 0.465
g/ml.Pattern recognition models using linear discriminant analysis(LDA)and k nearest neighbor(k-NN)classiers based on UV spectrum could successfully classify different types of injections and different manufacturers.The established method to measure tannin content based on UV spectroscopy is simple,rapid and reliable and provides technical support for quality control of tannin in Chinese medicine injections.

A modified resonant recognition model to predict protein-protein interaction

Proteins are fundamental components of all living cells and the protein-protein interaction plays an important role in vital movement.This paper briefly introduced the original Resonant Recognition Model(RRM),and then mod-ified it by using the wavelet transform to acquire the Modified Resonant Recognition Model(MRRM).The key characteris-tic of the new model is that it can predict directly the protein-protein interaction from the primary sequence,and the MRRM is more suitable than the RRM for this prediction.The results of numerical experiments show that the MRRM is effective for predicting the protein-protein interaction.

Molecular Dynamic Study for Chiral Discrimination of α-Phenylethylamine by Modified Cyclodextrin in Gas Chromatography

A molecular dynamic method in conjunction with a statistic test has been utilized to model chiral recognition of a-phenylethylamine on heptakis (2.6-di-O-butyl-3-O-butyryl)-β- cyclodextrin in gas chromatography. The modelling data correlated with the chromatographic elution order and indicated that the preferred site of α-phenylethylamine is the interior of cavity.

A study on continuous Chinese speech recognition based on stochastic trajectory models

After pointed the unreasonableness of the three basic assumptions contained in HMM, we introduce the theory and the advantage of Stochastic najectory Models (STMs) that possibly resolve these problems caused by HMM assumptions. In STM, the acoustic observations of an acoustic unit are represented as clusters of trajectories in a parameter space.The trajectories are modelled by mixture of probability density functions of random sequence of states. After analyzing the characteristics of Chinese speech, the acoustic units for continuous Chinese speech recognition based on STM are discussed and phone-like units are suggested. The performance of continuous Chinese speech recognition based on STM is studied on VINICS system. The experimental results prove the efficiency of STM and the consistency of phone-like units.

An auditory periphery model for improving narrow-band noise recognition rate of underwater targets

The recognition rate of the auditory periphery features decreases when the model is used to identify underwater targets in practice. To solve this problem, an improved method based on Gammatone filter bank is proposed. Firstly, after the reason of the decreasing of the recognition results is analyzed, the mechanism of multichannel data acquisition in acoustic engineering may narrow down signal frequency range, which leads to time-frequency features distortion. Secondly, the Gammatone filter bank is implemented to simulate frequency decom- position characteristics of human ear basilar membrane. Since the class information of the underwater noise signal is mostly contained in low frequency range, the auditory features of the conventional model are interpolated and the channel number of the filter bank and the central frequency of each frequency band are adjusted accordingly to obtain a 27-dimensional feature vector of the narrow-band target signal. The adjusted model may reflect the target＇s time- frequency feature more precisely. Finally, the performance of the auditory features is tested by a Neural Network classifier. The experiment results show that the modified auditory model is more effective than the conventional ones. The major information contained in broadband signals is reserved and the classification ability for real targets is further enhanced. The recog- nition results are increased from 82.59% to 88.80%. The modified auditory features effectively improve the recognition rate for underwater target radiated noise signals.

Spatiotemporal emotion recognition based on 3D time-frequency domain feature matrix

The research of emotion recognition based on electroencephalogram(EEG)signals often ignores the related information between the brain electrode channels and the contextual emotional information existing in EEG signals,which may contain important characteristics related to emotional states.Aiming at the above defects,a spatiotemporal emotion recognition method based on a 3-dimensional(3 D)time-frequency domain feature matrix was proposed.Specifically,the extracted time-frequency domain EEG features are first expressed as a 3 D matrix format according to the actual position of the cerebral cortex.Then,the input 3 D matrix is processed successively by multivariate convolutional neural network(MVCNN)and long short-term memory(LSTM)to classify the emotional state.Spatiotemporal emotion recognition method is evaluated on the DEAP data set,and achieved accuracy of 87.58%and 88.50%on arousal and valence dimensions respectively in binary classification tasks,as well as obtained accuracy of 84.58%in four class classification tasks.The experimental results show that 3 D matrix representation can represent emotional information more reasonably than two-dimensional(2 D).In addition,MVCNN and LSTM can utilize the spatial information of the electrode channels and the temporal context information of the EEG signal respectively.

Pilot protection of hybrid MMC DC grid based on active detection

Considering the advantages and limitations of traditional identification method,combined with the strategy of active detection,the principle of DC grid pilot protection based on active detection is proposed to improve the sensitivity and reliability of hybrid MMC DC grid protection,and to ensure the accurate identification of fault areas in DC grid.By using the DC fault ride-through control strategy of the hybrid sub-module MMC,the fault current at the converter station DC terminal is limited.Based on the high controllability of hybrid MMC,sinusoidal fault detection signals with the same frequency are injected into the line at each converter station.Based on model recognition,the capacitance model condition is satisfied by the detected signals at both ends during external faults whereas not satisfied during internal faults.The Spearman correlation coefficients is then introduced,and the correlation discriminant of capacitance model is constructed to realize fault area discrimination of DC grid.The simulation results show that the active detection protection scheme proposed in this paper can accurately identify the fault area of DC grid,and is not affected by fault impedance and has low sampling rate requirement.

A computer vision system for defect discrimination and grading in tomatoes using machine learning and image processing

With large-scale production and the need for high-quality tomatoes to meet consumer and market standards criteria,have led to the need for an inline,accurate,reliable grading system during the post-harvest process.This study introduced a tomato grading machine vision system based on RGB images.The proposed system performed calyx and stalk scar detection at an average accuracy of 0.9515 for both defected and healthy tomatoes by histogramthresholding based on themean g-r value of these regions of interest.Defected regionswere detected by an RBF-SVMclassifier using the LAB color-space pixel values.Themodel achieved an overall accuracy of 0.989 upon validation.Four grading categories recognitionmodelswere developed based on color and texture features.The RBF-SVMoutperformed all the explored modelswith the highest accuracy of 0.9709 for healthy and defected category.However,the grading accuracy decreased as the number of grading categories increased.A combination of color and texture features achieved the highest accuracy in all the grading categories in image features evaluation.This proposed system can be used as an inline tomato sorting tool to ensure that quality standards are adhered to and maintained.