Highest standard c
Highest standard count estimation
from fibre parameters using neural
network techniques
N Shanmugam & S S Doke
Received 19 April
2004; revised received 9 September 2004; accepted 10 November 2004
Artificial
neural network (ANN) model has been developed for predicting highest standard
count (HSC) from fibre properties, namely 2.5% span length, uniformity ratio,
micronaire and bundle strength. The developed ANN model was compared with the
multiple regression and fibre quality index (FQI) based regression models. ANN
ranking of fibre properties was carried out using difference in test
performance values as indicator and in case of multiple regression,
standardized regression coefficients were used. It has been observed that in
both ANN and multiple regression models, the ranks of span length and bundle strength
are the same. The span length is the largest contributor for HSC and the bundle
strength is the least contributor. The mean absolute errors of ANN and multiple
regression equation are found to be less by 15% and 11% respectively in
comparison with FQI-based linear regression equation.
Keywords: Artificial
neural network, Back propagation neural network, Cotton, Fibre quality index,
Highest standard count, Lea CSP, Multiple regression model
IPC Code: Int. Cl.7 D06H3/00,
G06N3/02