基于局部域超限学习机的石材识别算法  

A Stone Texture Classification Method Based on Local Receptive Field Extreme Learning Machine

作　　者：童冰[1] 许冲[2] 

机构地区：[1]漳州职业技术学院计算机工程系,福建漳州363000 [2]闽南师范大学计算机学院,福建漳州363000

出　　处：《闽南师范大学学报（自然科学版）》2016年第3期27-36,共10页Journal of Minnan Normal University：Natural Science

基　　金：福建省中青年科研项目资助(JA15687)

摘　　要：针对日益加快的石材加工及销售速度与缓慢的人工分选速度之间不协调导致的效率低下的问题,急需提出基于视觉的自动化石材识别算法.所提出的基于局部感受域超限学习机(Local receptive field extreme learning machine,LRF-ELM)的石材纹理识别算法包含图像预处理、ELM特征映射和ELM分类器三个模块.其中图像预处理模块首先要经过平滑滤波和图像增强操作降低噪声干扰,接着再进行白化操作,去除数据之间的相关性.相对于基于传统人工设计特征的方法,所提出算法的ELM特征映射模块通过ELM神经网络可自主学习出更接近高层语义的图像特征,显著提高分类的准确性.实验结果表明提出的算法相对于传统算法不仅具备较高的分类精度同时还具备较高的计算效率.Aiming at the problem of poor efficiency of stone production and sell caused by the mismatch between higher and higher speed of production and slow speed of artificial stone classification, it is necessary to put forward automatic computer vision based stone texture classification algorithm. This paper proposes a stone texture classification method based on local receptive field extreme learning machine （LRF-ELM） which consists of three modules：Image preprocessing, extreme learning machine feature mapping and a single classifier trained by ELM algorithm. The proposed method learns the feature by the ELM network itself and the features can be closer to the high-level semantics of images. On the image preprocessing module, smoothing filter and image enhancement operator are used to reduce the noise, and then whiten operator is used to Data decorrelation. Experimental results have shown that this proposed method obtains not only high classification accuracy but also extremely high computational efficiency comparing with other traditional algorithm.

关 键 词：石材纹理识别 局部感受域 白化 超限学习机 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]