Advantages of a hierarchical system of neural-networks for the interpretation of infrared spectra in structure determination

Cleva, Christophe; Cachet, Claude; Cabrol‐Bass, Daniel; Forrest, Tom

doi:10.1016/s0003-2670(97)00151-7

Cited by 19 publications

(24 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…I). These independent sets are the same as those used for a previous study on a hierarchical system of feedforward neural networks [8], allowing a comparison of the two approaches. The distribution criteria are used to check that the composition (proportion of compounds in each class) of the training and test sets is statistically representative of the composition of the whole database.…”

Section: Experiments Learning Parametersmentioning

confidence: 99%

“…The performance of the models based on winning units has been compared with those achieved by a hierarchical layered network which has been trained to recognise the same structural classes on the same learning set [8] two types of networks have a correlation coefficient of 0.94 leading to the conclusion that, despite their different architecture and classification method, they both rely on similar basis. This finding, already stated by Melsen et al [13], is reinforced by the localisation on the Kohonen map of the projections of the spectra of compounds from the learning set which have not been properly classified by the layered network.…”

Section: Comparison Of Model Performance With Those Achieved By a Laymentioning

confidence: 99%

“…The recent use of multilayered feed forward neural networks for the interpretation of spectral data shows good promise: the internal representations used by these networks are non-linear and are built up by a learning process based on examples. After the seminal works by Robb and Munk [2][3], many researchers explored the possibilities of these networks for the interpretation of infrared spectra [4][5][6][7][8]. However, the learning method used is a supervised learning process, and relies upon an existing structural classification.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Clustering of infrared spectra with Kohonen networks

Cleva

Cachet²,

Cabrol‐Bass³

1999

Analusis

View full text Add to dashboard Cite

Abstract. The design of systems for spectral data interpretation requires clustering of chemical compounds based on their spectral characteristics. Kohonen networks have been shown to be efficient tools to achieve this clustering. These auto-organising systems perform a mapping between a high-dimensional variable space and a two-dimensional one. An application to infrared spectra of organic compounds is presented here. The non-supervised learning algorithm used allows classification of compounds by spectral characteristics without a priori knowledge. An analysis of the distribution of spectra on the resulting maps is used to build models for predicting the presence or absence of specific structural features. The performance of the models in recognising structural features is discussed and compared with the prediction of a multilayered feed forward network (MLFFN). Localisation of compounds wrongly classified by the MLFFN on the Kohonen maps allows to establish a link between the supervised and the unsupervised approaches.

show abstract

Section: Experiments Learning Parametersmentioning

confidence: 99%

Section: Comparison Of Model Performance With Those Achieved By a Laymentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Clustering of infrared spectra with Kohonen networks

Cleva

Cachet²,

Cabrol‐Bass³

1999

Analusis

View full text Add to dashboard Cite

show abstract

“…There are a number of established techniques for exploiting spectral content of MS imagery for remote sensing applications (Tassel Cap [Kauth and Thomas, 1976], [Crist and Cicone, 1984]; Atmospherically Resistant Vegetation Index [Kaufman and Tanre, 1992]; Normalized Difference Vegetation Index [NDVI] [Goward, Markham, Dye, Dulaney, and Yang, 1991]; abductive polynomials [Drake, Kim, and Kim, 1993]; and principal component analysis (PCA) [Jiaju, 1988], [Cleva, Cachet, Cabrol-Bass, and Forrest, 1997], to name a few). However, when these techniques are applied to HS data, there are several difficulties.…”

Section: Multi-band Analytical Techniquesmentioning

confidence: 99%

Hyperspectral and Spectro-Polarimetric Pixel-Level Classification Using Genetic Programming

Rauss¹

2001

View full text Add to dashboard Cite

The objective force will be relying heavily on their sensors to be a combat multiplier to help improve the force's effectiveness and survivability, particularly for reconnaissance, surveillance, and target acquisition missions. Currently, fielded passive sensor systems are generally ineffective against camouflage, concealment, and deception. Their performance is also sensitive to environmental conditions. To meet future needs, several new sensor systems are being developed and evaluated. Two of these new sensors are passive systems that collect additional, measurable characteristics of light: hyperspectral (HS) systems and spectro-polarimetric (SP) systems.To fully take advantage of the information that these systems collect requires new algorithms and techniques. This report discusses why new techniques are necessary and details the development of a computer-assisted design system for the discovery of classification algorithms via a small number of sample target and background signatures. The technique is called genetic programming (GP). GP is an adaptive learning technique that automatically generates a computer program (in this work, a mathematical equation) to solve the problem it is given. This report documents work conducted primarily between September 1999 and August 2000, while the author was on a rotation at the University of Michigan under the Federated Laboratories Consortium program. The report demonstrates that GP could be a useful technique for processing HS and SP data. The experiments reported here show that by using even the simplest of operators (addition, subtraction, multiplication and division) the GP process can develop interesting and potentially useful solution equations. The results shown here are encouraging. However, many questions remain to be answered.

show abstract

“…Although it reflected the absorption rules of spectral-structural feature, it is still not sufficient. Along with the development of chemometric tools in infrared spectra, it has boosted many new methods in automated machine interpretation, such as artificial neural networks (ANN) [6][7][8], partial least squares [9,10] and support vector machine (SVM) [11][12][13]. These computational methods have brought new insights into the spectroscopic analysis.…”

Section: Introductionmentioning

confidence: 99%