Identifying defects on plywood using a minimum distance classifier and a neural network
M.S. Packianathera, P.R. Drakeb
aSchool of Engineering, Cardiff University, UK
bUniversity of Liverpool, UK
Contents
-
Introduction
-
A generic Classifier
-
The minimum distance classifier (MDC)
-
Calculating the weights of the MDC
-
Neural Network (NN)
-
Calculating the weights of the NN
-
Wood veneer inspection application
-
Training the classifiers
-
Comparison of WVIMDC and WVINN
-
Conclusion
Introduction
-
Two types of classifiers considered
-
The minimum distance classifier
-
a relatively simple classifier
-
The neural network
-
an intelligent classifier
-
The application considered
-
identifying defects on plywood
-
Wood veneer inspection minimum distance classifier (WVIMDC)
-
Wood veneer inspection neural network (WVINN)
Training the classifiers
For the application studied here, 232 examples (defects and clear wood) were employed. The classification of these examples had been performed by a human inspector. For subsequent classification experiments, for each class, 80% (185 in total) of the examples were selected at random to form the training set and the remaining 20% (47 in total) formed the test set. Three such sets of training and corresponding test data were created and are referred to as RS1, RS2 and RS3 in this paper.
Conclusion
-
The performance of the two classifiers, namely, WVIMDC and WVINN has been compared for identifying defects on plywood.
-
The results show that the efficiency of the WVINN is higher than that of the WVIMDC.
-
This shows that the neural network behaves like a non-linear machine by having more than one layer of weights, and can be trained to learn the non-linear discriminating functions between the different classes.
