基于多层感知机的木材颜色分类

Wood Color Classification Based on Multi-layer Perception

目前的木材分类大多需要依靠人工挑选方式完成,但人工分类受操作人员主观因素、工作经验、劳动强度等因素的制约,已经很难适应产业的发展。为了有效提高木材颜色分类的自动化程度及分类的准确性,采用机器视觉技术对木材表面颜色分类方法展开研究。通过使用python opencv软件对木材表面进行高斯滤波、滚动引导滤波以及特征提取预处理操作,实现对木材图像的去背景化,并将图像的RGB、Lab低阶矩作为颜色分类的特征,最后采用多层感知机(MLP)构建木材表面颜色的分类器。选取了320张无损实木板材的完整表面图像作为训练样本,其中25%为测试集,75%为训练集。结果表明:该分类器对木材表面颜色分类的准确率为96.25%,验证了将多层感知机模型应用在木材颜色分类方面的有效性。

Most of the current wood classification needs to be completed by manual selection,but manual classification is constrained by factors such as workers'subjective factors,work experience and labor intensity,making it difficult to adapt to the development of the industry.In order to effectively improve the classification accuracy of the degree of automation for wood color classification,a machine vision technology was adopted in this paper to do research,with python opencv software specifically used to achieve Gaussian filtering,rolling guide filtering and feature extraction pre-processing operations on the wood surface,realizing the de-background of wood images,with RGB,Lab low-order moments of images used as the color classification feature.Finally,a multilayer perceptron(MLP)was used to construct a classifier for wood surface colors.320 complete surface images of non-destructive solid wood panels were selected as training samples,of which 25%was the test set and 75%was the training set.The results show 96.25%accuracy rate of the classifier for color classification of wood surface,thereby verifying effectiveness of the layer perceptron model in wood color classification.