一种改进的尺度不变特征识别算法用于建盏釉面的相似性判定北大核心CSCD

为解决建盏真伪区分问题,通过改进的尺度不变特征识别算法,建立实验原型系统,用于识别不同类型釉面的建盏。研究的识别模型在云端上提供服务,以便对建盏釉面的相似性进行判断。该系统采用Streamlit平台构建实验性人机交互界面。针对识别模型,研究开发了基于边界跟随法的图像预处理策略,旨在滤除背景和无关材料特征干扰,提高自然条件下拍摄的图像识别能力,以减少对建盏釉面的误判可能性。利用改进的基于支持向量机(support vector machines,SVM)的尺度不变特征变换(scale-invariant feature transform,SIFT)特征匹配分类策略,设计了多项性能实验,以获得足够的数据集分辨率、SVM最优超参数和初始化所需的最小数据集参数模型。经实验证实,系统可清楚地区分相同类型釉面图案(兔毛釉、滴油釉)的相似性。它使用给定的有限数据集提供高达92.60%的识别准确率,并将单次识别速度提高0.84 s。

An improved scale-invariant feature recognition algorithm for similarity judgme nt of the glaze of Jianzhan

In order to solve the problem of distinguishing the authenticity of Ji anzhan,an experimental prototype system is established through an improved scale-invariant feature re cognition algorithm to identify different types of the glaze of Jianzhan.The researched recognition mo del provides services on the cloud in order to judge the similarity of the glaze of Jianzhan.The system uses the Streamlit platform to build an experimental human-computer interaction interface.For the recognit ion model, an image preprocessing strategy is researched and developed based on the boundary following method,which aims to filter out the interference of background and irrelevant material characteristics,impr ove the recognition ability of images taken under natural conditions,and reduce the possibility of misjudgm ent of the glaze of Jianzhan.Using an improved support vector machine (SVM)-based scale invariant feature transform (SIFT) feature matching classification strategy,a number of performance experim ents are designed to obtain sufficient data set resolution,SVM optimal hyperparameters and initializ ation required minimum data set parameter model.Experiments have confirmed that the system can clearly distinguish the similarities of the same types of glaze patterns (rabbit hair glaze,oil drip gl aze).It uses a given limited data set to provide a recognition accuracy rate of up to 92.60% and increases th e single recognition speed by 0.84 s.