基于E-nose的多类饮料识别：改进的编码Fisher判别分析方法
172 浏览量
更新于2024-08-26
收藏 640KB PDF 举报
本文主要探讨了"改进的Fisher判别分析方法在电子鼻（Electronic Nose, E-nose）支持下的饮料识别"这一主题。电子鼻作为食品工业中的重要工具，特别是在饮料识别领域展现出其显著的应用价值。传统Fisher判别分析（FDA）作为一种二分类器，其主要优势在于其简便的模式识别能力，然而在实际应用中，尤其是日常生活中常见的多类别识别问题上，其局限性就显得尤为明显。
为了克服FDA在多分类任务中的不足，作者提出了一种基于编码的Fisher判别分析（Encoding Fisher Discriminate Analysis, EFDA）方法。这种方法旨在扩展FDA的适用范围，使其能够处理复杂的多类识别问题。研究者选取了七种常见的饮料样本，包括中国白酒、啤酒、葡萄酒、绿茶、红茶、乌龙茶和普洱茶，通过电子鼻进行客观测试，以此验证EFDA在多类别饮料识别中的性能和有效性。
在改进的EFDA方法中，可能包括对原始数据的预处理步骤，如特征选择和降维，以提高分类器的准确性和效率。同时，可能会采用编码策略将多类问题转化为一系列二分类问题，使得FDA的核心思想得以应用。此外，可能还涉及到训练过程中的参数优化和模型评估，以确保EFDA在实际应用中的鲁棒性和稳定性。
该研究不仅关注技术层面的算法创新，还考虑到了实用性，即如何使电子鼻在饮料识别场景中发挥更大的作用，这对于食品质量控制、生产监控以及消费者体验提升等方面具有重要意义。通过实验结果的展示，读者可以了解到EFDA在饮料分类中的精确度和性能提升，这为未来电子鼻在更多领域的多分类问题提供了新的思路和技术参考。
这篇文章深入探讨了如何通过改进的Fisher判别分析方法来解决电子鼻在多类别饮料识别中的挑战，展示了理论与实践的结合，并为相关领域的研究人员和工程师提供了实用的工具和方法论指导。
Electronic Nose Based Beverage Identification by an Improved Fisher Discriminate
Analysis Method
Tao Liu, Lei Zhang, Yanbin Chen, Dongqi Li, Mengya Wu and Xingrui Cui
College of Communication Engineering, Chongqing University, Chongqing 400044, China
cquliutao@cqu.edu.cn; leizhang@cqu.edu.cn
Abstract—Electronic noses are widely employed in food
industry and play an important role in beverage
identifications. Meanwhile, fisher discriminate analysis is
considered as one of convenient pattern recognition
methodologies for electronic noses. Traditional fisher
discriminate analysis (FDA) is a kind of binary classifiers. In
other words, it is not competent to multi-class classifications
which are common in daily concerns. To solve the limitation of
the binary classifier, we presented a FDA based classification
method, namely encoding fisher discriminate analysis (EFDA),
which can be applied to multi-class recognition. In this paper,
seven kinds of beverages including Chinese liquor, beer, wine,
green tea, black tea, oolong tea and pu’er tea were prepared
and tested objectively by an electronic nose (E-nose) system
with 32 gas sensors. The E-nose system was a prototype for
general detection of complicated gaseous compounds . It
performed all the experiments and tested the volatile
compounds of seven beverages. The experimental results
indicated that the proposed method were more effective than
conventional pattern recognition methods such as principal
component analysis (PCA) and k-nearest neighbor (k-NN).
The highest recognition rate could reach 96.83% by EFDA.
The outcomes achieved demonstrated that EFDA can be
successfully used in multi-class identifications.
Keywords—electronic nose; fisher discriminate analysis;
beverage; binary encoding
I. I
NTRODUCTION
Artificial olfactory sensation is a novel technology to
simulate human-smell process to indentify various gaseous
compounds. Typically, when a gas sensor array associates
with pattern recognition models, an artificial olfactory
system is composed. This kind of machine, namely
electronic nose (E-nose), has great potential in complicated-
ingredient-gas sensing for their evident superiorities such as
high speed, good reproducibility and non-invasion.[1] On
the other hand, food safety and quality become more and
more important in modern society due to increasing concern
on human health and living quality. Thus, instead of
traditional chemical instruments, E-noses supply a
convenient way to perform special detections of food
industry in which beverage discriminations usually play an
important role. [2]
Communities have made quite a few efforts to the
scientific researches about beverage identifications. Among
them, some scholars pay attention to distinguish various
categories of tea according to types, grades, even habitats
with commercial E-nose systems.[3]-[5] Meanwhile, some
studies focus on differentiating kinds of alcoholic drinks on
their different aromatic flavors.[6][7] Furthermore, positive
attempts have been performed to show strong ability of
olfactory machines to recognize different fruit juices and
milks. [8]-[10] Generally speaking, suited pattern
recognition models are needed to infer certain results from
sensor-array responses in E-nose machines. As far as it is
concerned, a few methods such as principal component
analysis (PCA), fisher discriminate analysis (FDA), k-
nearest neighbor algorithm (k-NN) and artificial neural
network (ANN) have been adopted for E-nose based
beverage tests and shown obvious
validity and practicability in the reports aforementioned.
Although FDA with the advantages of simple and easy
realization is frequently used in pattern recognition of E-
nose systems, conventional FDA method only fits for a
binary classification.[11]-[13] That is to say that FDA is
only a binary classifier. In order to extend the applied range
of FDA to multi-class identifications, either one-against-
one or one-against-all strategies[14][15] has been
considered to be a suitable scheme instead of the traditional
one. However multi-class identification of FDA, even other
binary classifiers, is so far an interesting issue for the need
of convenient recognition solutions. The wide applications
and issues of E-nose can be referred to as [16]-[18].
In this paper, the authors proposed a novel multi-type
classification strategy, called Encoding FDA (EFDA),
which combined binary encoding and hamming-distance
measurement with FDA to explore a better FDA based
multi-class classifier. Meanwhile, we also designed and
developed an E-nose platform with 32 gas sensors for gas
tests. Seven types of beverages including Chinese liquor,
beer, wine, green tea, black tea, oolong tea and pu’er tea
were prepared as testing samples. Then EFDA, k-NN and
PCA were adopted as pattern recognition part of E-nose.
Comparison of the three method was presented and the
recognition accuracy of EFDA was the best one among
three adopted methods. It hinted that EFDA had favorable
prospects in multi-class applications. The rest of this paper
is organized as follows: In Section II, we illustrate the
hardware architecture of the proposed E-Nose platform and
describe the principle of the methods used in algorithm
validation. Then the experimental setup, results and
2016 Asia Multi Conference on Modelling and Simulation
2376-1172/16 $31.00 © 2016 IEEE
DOI 10.1109/AMC.2016.20
77
下载后可阅读完整内容，剩余5页未读，立即下载
点击了解资源详情
592 浏览量
点击了解资源详情
2021-09-15 上传
263 浏览量
332 浏览量
2021-09-23 上传
135 浏览量
2021-09-23 上传
