融合HOG+SVM与MFCC特征的自动优化物体识别系统研究

Self-optimizing Object Recognition Based on Improved HOG+SVM and MFCC Features

在与人交互情况下,针对物体识别系统通过反馈信息自动优化识别能力问题,提出一种结合梯度直方图(HOG)特征提取和支持向量机(SVM)进行特定物体识别的方法。运用Tensorflow训练语音识别模型反馈人机交互信息,使系统实现自优化。以手表类别作为识别对象,通过HOG特征描述对手表进行特征提取,通过二维主成分分析(2DPCA)和线性判别分析(LDA)对整体和局部特征进行降维,运用改进的空间金字塔匹配模型通过SVM对其分类,并运用非极大值抑制(NMS)确定区域,运用训练的梅尔倒谱(MFCC)特征语音识别模型对反馈信息进行识别,最终整合信息优化识别系统。实验表明,该系统对手表有较高的识别率,并能通过人机交流在较短时间内使系统实现自优化。

Aiming at the problem that the object recognition system automatically optimizes the recognition ability through feedback information in the case of interaction with humans,a gradient histogram(HOG)feature extraction and support vector machine(SVM)for specific object recognition are proposed.Meanwhile,the Tensorflow speech recognition model feeds back the human-computer interaction information and then enables the system to implement a self-optimization method.The method uses the watch as the recognition object,and extracts the features of the watch through the HOG feature descriptor.The two-dimensional principal component analysis(2DPCA)and linear discriminant analysis(LDA)are used to reduce the overall and local features,and the improved spatial pyramid is used.The matching model is classified by SVM,the region is confirmed by non maximum suppression(NMS)and the feedback recognition information is identified by the self-trained speech recognition model using Mel-Frequency cepstral coefficients(MFCC)features.Finally,the identified information optimization recognition system is integrated.Experiments show that the system has a high recognition rate for watches,and it can realize self-optimization of the system in a short time by human-machine communication.