差分振子相图的自动识别与应用

通过仿真实验证明了差分振子相图的大小和待检测信号幅值大小之间的关系。在相同的参数条件下,差分振子的相图越大,则待检测信号的幅值越大。因此,在参数相同的条件下,对不同的差分振子相图进行比较,可以得到待检测信号幅值间的大小关系。针对差分振子相图的特点给出了差分振子识别的新方法,可以快速识别差分振子是收敛于极环还是极点。最后,通过工程数据的分析成功发现了设备故障发生发展的劣化过程,并对差分振子的相图进行识别,取得了理想的效果。

Recognition of Similar Weather Scenarios in Terminal Area Based on Contrastive Learning

In order to improve the recognition accuracy of similar weather scenarios(SWSs)in terminal area,a recognition model for SWS based on contrastive learning(SWS-CL)is proposed.Firstly,a data augmentation method is designed to improve the number and quality of weather scenarios samples according to the characteristics of convective weather images.Secondly,in the pre-trained recognition model of SWS-CL,a loss function is formulated to minimize the distance between the anchor and positive samples,and maximize the distance between the anchor and the negative samples in the latent space.Finally,the pre-trained SWS-CL model is fine-tuned with labeled samples to improve the recognition accuracy of SWS.The comparative experiments on the weather images of Guangzhou terminal area show that the proposed data augmentation method can effectively improve the quality of weather image dataset,and the proposed SWS-CL model can achieve satisfactory recognition accuracy.It is also verified that the fine-tuned SWS-CL model has obvious advantages in datasets with sparse labels.

Action Recognition from a Different View

In this paper,we propose a novel approach to recognise human activities from a different view.Although appearance-based recognition methods have been shown to be unsuitable for action recognition for varying views,there must be some regularity among the same action sequences of different views.Selfsimilarity matrices appear to be relative stable across views.However,the ability to effectively realise this stability is a problem.In this paper,we extract the shape-flow descriptor as the low-level feature and then choose the same number of key frames from the action sequences.Self-similarity matrices are obtained by computing the similarity between any pair of the key frames.The diagonal features of the similarity matrices are extracted as the highlevel feature representation of the action sequence and Support Vector Machines(SVM) is employed for classification.We test our approach on the IXMAS multi-view data set.The proposed approach is simple but effective when compared with other algorithms.

Statistics and Its Applications in Identification of Irregular Particles

Phase Doppler anemometry(PDA) is very sensitive to the shape of testing particles, which is based on sphericity assumption and Mie’s theory. In practice, there exists effectiveness of non sphericity and the response of PDA system deviates from the theoretical prediction. In this paper, the statistic characteristics of PDA signal are analyzed and a method of identifying and quantifying irregular particles is proposed. It is concluded that phase difference of PDA signal for irregular particles is an unbiased estimation for spherical particles.

Automatic recognition and quantitative analysis of Ω phases in Al-Cu-Mg-Ag alloy

The main methods of the second phase quantitative analysis in current material science researches are manual recognition and extracting by using software such as Image Tool and Nano Measurer. The weaknesses such as high labor intensity and low accuracy statistic results exist in these methods. In order to overcome the shortcomings of the current methods, the Ω phase in A1-Cu-Mg-Ag alloy is taken as the research object and an algorithm based on the digital image processing and pattern recognition is proposed and implemented to do the A1 alloy TEM （transmission electron microscope） digital images process and recognize and extract the information of the second phase in the result image automatically. The top-hat transformation of the mathematical morphology, as well as several imaging processing technologies has been used in the proposed algorithm. Thereinto, top-hat transformation is used for elimination of asymmetric illumination and doing Multi-layer filtering to segment Ω phase in the TEM image. The testing results are satisfied, which indicate that the Ω phase with unclear boundary or small size can be recognized by using this method. The omission of these two kinds of Ω phase can be avoided or significantly reduced. More Ω phases would be recognized （growing rate minimum to 2% and maximum to 400% in samples）, accuracy of recognition and statistics results would be greatly improved by using this method. And the manual error can be eliminated. The procedure recognizing and making quantitative analysis of information in this method is automatically completed by the software. It can process one image, including recognition and quantitative analysis in 30 min, but the manual method such as using Image Tool or Nano Measurer need 2 h or more. The labor intensity is effectively reduced and the working efficiency is greatly improved.

Fingerspelling Recognition by Hand Shape Using Higher-Order Local Auto-Correlation Features

The fingerspelling recognition by hand shape is an important step for developing a human-computer interaction system. A method of fingerspelling recognition by hand shape using HLAC （higher-order local auto-correlation） features is proposed. Furthermore, in order to use HLAC features more effectively, the use of image processing techniques： reducing an image resolution, dividing an image, and image pre-processing techniques, is also proposed. The experimental results show that the proposed method is promising.

Inverse design of an integrated-nanophotonics optical neural network

Artificial neural networks have dramatically improved the performance of many machine-learning applications such as image recognition and natural language processing. However, the electronic hardware implementations of the above-mentioned tasks are facing performance ceiling because Moore’s Law is slowing down. In this article, we propose an optical neural network architecture based on optical scattering units to implement deep learning tasks with fast speed, low power consumption and small footprint.The optical scattering units allow light to scatter back and forward within a small region and can be optimized through an inverse design method. The optical scattering units can implement high-precision stochastic matrix multiplication with mean squared error < 10-4 and a mere 4*4 um2 footprint.Furthermore, an optical neural network framework based on optical scattering units is constructed by introducing "Kernel Matrix", which can achieve 97.1% accuracy on the classic image classification dataset MNIST.