基于融合特征以及卷积神经网络的环境声音分类系统研究

环境声音识别系统主要基于深度神经网络以及种类繁多的听觉特征对环境声音进行分类识别。分析基于深度神经网络的环境分类任务中,哪种听觉特征更适合环境声音识别系统十分必要。选择了基于2个广泛使用的滤波器:梅尔和Gammatone滤波器组提取的3种声音特征。随后,提出了一个MFCC和GFCC融合的特征MGCC。最后采用文中提出的深度卷积神经网络来验证哪种特征更适合于环境声音的分类识别。实验结果表明,在基于神经网络的环境声音分类系统中,信号处理特征比频谱图特征的效果好,其中,MGCC特征具有比其他特征更好的性能。最后,用文中提出的MCC-CNN模型与其他环境声音分类模型在UrbanSound 8K数据集上进行了对比。实验结果表明,所提模型分类精度最好。

符号延异与图像缺失:明代山水画中的寻道者意象与艺术主流的终结

有明一代山水画发展的主流,经历了一个由外向内的转变,前期画家寻道于自然,后期画家转而寻道于内心。本文通过图像研究而旨在阐述:明代山水画主流的转向,实为对前代山水画发展主流的重复,而重复的终点则是明末山水画中点景小人的缺席,这标志了主流的终结。这一探讨从德里达关于意指延异的概念出发,借鉴解构主义的双重阅读,来分析明代山水画中寻道者的意象及其在符号系统中作为意指中心的功能。本文结论是,当文人山水画发展到明末,寻道者意象的缺失,使中国山水画的意指系统被解构,艺术史的主流遂于明末消散。

AUTOMATIC ORDER-DETERMINATION OF TRANSIENT IN--PLANE DISPLACEMENT MOIRE PATTERNS

This paper presents a fringe-carrier method that eliminates sign ambiguity of transient moire fringes which can be used to automatically determine the relative orders. A fringe carrier is preset in the static state of the specimen and the dynamic in-plane displacements are recorded as the modulation to the frequency of the carrier fringes when the specimen is loaded by impact. According to a modulating criterion developed from the modulation degree, the fringes of the transient moire patterns keep monotonical in orders so that they can be automatically encoded in grey levels by a digital image system. The moire orders purely caused by dynamic loadings are evaluated by subtracting the grey-value of the unmodulated carrier image from that of the modulated carrier images encoded by their orders. With the subtracted moire orders the strain components can be obtained, and, correspondingly, the histograms of dynamic displacement moire images are shown with order variation by image-difference.