Classification by varying features with an erring sensor

doi:10.1016/0005-1098(94)90054-X

Automatica

Volume 30, Issue 12, December 1994, Pages 1943-1948

https://doi.org/10.1016/0005-1098(94)90054-X Get rights and content

Abstract

A method is proposed for unsupervised classification by a feature that may vary with time, measured by an erring sensor. A classification threshold for the erring sensor is found such that the misclassification is minimized. It is shown that the method is an application of Bayes rule without knowledge of the a priori probabilities, while estimating the class conditional probabilities by an erring sensor model. Sorting of fruits is presented as an illustrative example.

References (19)

U. Ben-Hanan et al.
Classification of fruits by a Boltzmann perceptron neural network
Automatica
(1992)
J.O. Berger
R.E. Boucher et al.
Adaptive detection and removal of non-Gaussian spikes from Gaussian data
IEEE Trans. Pattern Anal. Machine Intell.
(1982)
K.A. Dines et al.
Constrained least squares filtering
IEEE Trans. Acoust. Speech Signal Processing
(1977)
R.O. Duda et al.
I. Foroutan et al.
Feature selection for automatic classification of non-Gaussian data
IEEE Trans. Syst., Man, Cybern.
(1987)
K. Fukunaga
V. Hasselblad
Estimation of parameters for a mixture of normal distributions
Technometrics
(1965)
B.R. Hunt
Deconvolution of linear systems by constrained regression and its relationship to the Wiener theory
IEEE Trans. Autom. Control
(1972)

There are more references available in the full text version of this article.

Cited by (1)

Adaptive classification - A case study on sorting dates
2000, Journal of Agricultural and Engineering Research
The probability of classifier errors in automated grading of fruits is much greater than in traditional well-defined and highly separated, static classification tasks. Presently, operators of conventional sizers and colour sorters adjust the class boundaries manually based on observations of obvious misclassification trends in the packed fruit, with the goal of minimizing the classification errors. However, the new sorting machines utilize many features to reach the grade decision. A human operator is unable to control the multitude of parameters under control. Estimating the between-class discriminant functions requires estimation of the a priori class probabilities (‘priors’) and the class-conditional probability densities. The time-varying nature of the priors and the probability densities result in unsatisfactory classifier performance. To solve these problems, an adaptive grading approach by ‘prototype populations’ is proposed. The produced stream is classified into a discrete number of prototype streams or populations by a global ‘population classifier’. For each unique prototype population a separate, optimal ‘grade classifier’ is designed for sorting individual fruits. The global ‘population classifier’ utilizes a finite-length stack of features continuously updated from the most recently sorted produce. The statistical attributes of the features sample in the stack are analysed to determine which produce population is currently passing through the system. When the population classifier determines that the stack contents have originated from a different prototype population, it changes the active ‘grade classifier’ to the most appropriate one for the current fruit population. An example of simulated adaptive versus conventional train-once, sort-many, grading is presented on data sets obtained from a system to sort dates by machine vision. The example demonstrates that adaptive grading by prototype populations yields lower misclassification rates in comparison to conventional sorting.

^☆: This paper was not presented at any IFAC meeting. This paper was recommended for publication in revised form by Editor A. Sage.

View full text

Automatica

Brief paperClassification by varying features with an erring sensor☆

Abstract

Classification of fruits by a Boltzmann perceptron neural network

Automatica

Adaptive detection and removal of non-Gaussian spikes from Gaussian data

IEEE Trans. Pattern Anal. Machine Intell.

Constrained least squares filtering

IEEE Trans. Acoust. Speech Signal Processing

Feature selection for automatic classification of non-Gaussian data

IEEE Trans. Syst., Man, Cybern.

Estimation of parameters for a mixture of normal distributions

Technometrics

Deconvolution of linear systems by constrained regression and its relationship to the Wiener theory

IEEE Trans. Autom. Control

Adaptive classification - A case study on sorting dates

Brief paper
Classification by varying features with an erring sensor☆