V.T.Grinchenko, V.V.Krizhanovskii, V.V.Krizhanovskii (jr.) Algorithms for adaptive and rank classification of the breath sounds |
Acoustic bulletin, Vol. 5 ¹ 3, (2002) |
| Originating from statistical approach the problem of classification of the state of human respiratory tract is formulated. At that the power spectral density of the breath sounds is used as the basic informational characteristic. To minimize an a priori information, the case of separation of spectral characteristics into two classes is considered. These classes cover the persons with healthy and those with pathological respiratory tracts. Using a training data sample of the breath sounds from healthy persons the algorithm of adaptive classification is synthesized. Analysis of the algorithm's structure is conducted and its simplification is carried out in order to increase the algorithm's stability to amplitude coefficients being insignificant for the classification problem. The problem of classification at absence of the training breath sounds data is considered. The algorithms of classification using the information on structure of ranks of the spectral power density are offered. It is shown that mentioned techniques are not sensitive to the amplitude characteristics accounting the amplification factors and a rhythmicity of respiration. The developed algorithms are checked experimentally. Conditions for increase of their efficiency and reliability are determined. |
|
KEY WORDS: breath sounds, signal processing, algorithm, adaptive classification, rank classification |
| TEXT LANGUAGE: Russian |