A First Course in FUZZY and NEURAL CONTROL

8.4. IDENTIFICATION OF TRASH IN COTTON 273

Table 8.7. Final values of consequent

parameters: 232-partition

Consequent Parametersaji

ji 01 2 3

15. 6045 − 0. 0623 5 .4067 10. 2944

20 .0746 0. 2576 0. 2428 − 5. 9242

3 − 1 .8363 0. 0098 1. 3515 − 0. 7228

40. 1011 − 0. 0361 0 .4841 2. 0547

51. 2008 − 0. 0002 − 0. 3727 − 0. 4426

60 .7795 0. 0028 0. 3548 − 1. 0790

7 − 3 .5730 0. 0003 − 1 .9937 4. 0406

86 .4012 0. 0003 − 28 .0326 27. 2515

95. 9735 − 0. 0002 − 4. 8844 − 3. 2546

10 2. 1531 − 0. 0001 1. 4765 − 4. 9520

11 − 1 .4469 0. 0002 2 .1095 2. 7445

12 − 0 .2404 0. 0001 0 .0892 1. 3792

Table 8.8. Initial andfinal values of the premise parameters: 232-partition.

Ling. Var. Area Solidity Edif

Ling. Val. Small Large Small Medium Large Small Large

Initial a 180.0000 4820.0000 0.3333 0.3333 0.3333 0.2500 0.2500

Para- b 2.0000 40.0000 2.0000 2.0000 2.0000 2.0000 2.0000

meters c 0.0000 5000.0000 0.0000 0.5000 1.0000 0.0000 0.5000

Final a 179.9964 4819.9983 0.4730 0.4839 0.0441 0.1097 0.0874

Para- b 2.2446 40.0018 1.9899 2.0003 2.0442 2.0230 2.0453

meters c -0.0010 5000.0017 0.0983 0.5528 0.9362 0.0057 0.3249

Figures 8.15ó8.17 show the initial and thefinal (tuned) membership func-
tions forarea,solidity,andEdifmeasures. Based on the premise and consequent
parameters, the training data is classified to evaluate the classification results.

The decision procedure for the identificationofthetrashtypesisasfollows:
The input pattern for each object in the test sample is passed in the forward
direction. With thefinal premise parameters defining the membership functions,
the outputs at each layer are calculated. The outputy which is the sum of the
outputs of Layer 4 is calculated. The outputs of the adaptive nodes at Layer
4 use the updated consequent parameters. Based on the outputy, the input
test pattern is classified to belong to a particular trash type. If the output is in
the range of the target value±0.1, then the trash type is classified to belong to
that trash type. For example, if the output of the network lies in the range of
0.2±0.1, it is classified as bark, since 0.2 is the desired value of bark objects.
For example, the output for object 16 (pattern 9) of test sample #1 (TA114) is
0.2777 and is classified as a bark object (Table 8.10).