织物摩擦声音的特征提取与辨别  

Feature extraction and discrimination of fabric frictional sounds

作　　者：杨剑平[1,2] 孟旭 徐蓉蓉[1] 陶晨[1] YANG Jianping;MENG Xu;XU Rongrong;TAO Chen(College of Textile and Garment,Shaoxing University,Shaoxing,Zhejiang 312000,China;Key Laboratory of Clean Dyeing and Finishing Technology of Zhejiang Province,Shaoxing,Zhejiang 312000,China)

机构地区：[1]绍兴文理学院纺织服装学院,浙江绍兴312000 [2]浙江省清洁染整技术研究重点实验室,浙江绍兴312000

出　　处：《毛纺科技》2020年第3期21-26,共6页Wool Textile Journal

基　　金：国家自然科学基金青年项目(51703130)。

摘　　要：针对不同织物摩擦声音的辨别问题,本文在哈尔特征基础上构建辨别机制,通过不同种类织物摩擦声音的差异,探索高效的织物摩擦声音辨别方法。通过摩擦声音采集装置采集了4组织物摩擦声音信号,对采集到的信号提取所有尺度及位移上的哈尔特征来构建1个特征空间;在特征空间中对每个点建立1个相应的辨别器,该辨别器允准该类所有正样本尽可能多地否决该类负样本;利用单个负样本被大量辨别器允准的概率非常小这一特点过滤掉该类全部负样本;对辨别器进行渐进性选择以构建辨别器队列,得到了具有同等辨别力的无序辨别器精简版本。结果表明:本文构建的辨别器队列可在大幅减少辨别器数量的同时维持织物摩擦声音辨别的正确率不变。To discriminate different fabric frictional sound, a mechanism based on Haar features was established in this study, and an efficient approach was conducted to identify fabric friction sounds by discriminating against the performances of different fiber materials manifested in the mechanism. 4 categories of fabrics were put through a specialized apparatus to collect frictional sound signals. The Haar features on every scale and position of the acquired signal were extracted so as to establish a feature space. For each point in the feature space, a discriminator was built to approve all positive samples of a certain category and deny negative samples as many as possible. All negatives were filter out due to the probability of a single negative sample being allowed by a large number of discriminators is very small. The progressive selection was performed on the discriminators to form a queue so as to bring out a much reduced version of the unordered discriminators with the same discriminability. Results have indicated that the discrimination queue suggested by this study can substantially reduce the number of discriminators involved and meanwhile keep the discriminability of fabric friction sounds invariable.

关 键 词：织物摩擦声音 哈尔特征 负样本 辨别器队列 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]