278 CHAPTER 8. APPLICATIONS
Table 8.15 illustrates the classification results of the training data. As seen
from the table, the ANFIS performance is excellent in identifying the trash
types. Of the 18 bark objects, 2 objects were misclassified, one as stick and one
as leaf. Of the 34 stick objects, 2 were classified as leaf. Only 1 object out of 54
leaf objects was classified as stick. Similarly, only 1 object is classified as leaf
out of the 107 pepper objects. This classification rate with the 222-membership
partition is 97.1831%.
Table 8.14. Initial andfinal values of the premise parameters: 222-partition
Linguistic variable Area Solidity Edif
Linguistic value Small Large Small Large Small Large
a 160.0000 4840.0000 0.5000 0.5000 0.2500 0.2500
Initial b 2.0000 60.0000 2.0000 2.0000 2.0000 2.0000
parameters c 0.0000 5000.0000 0.0000 1.0000 0.0000 0.5000
a 160.0040 4839.9914 0.6568 0.3598 0.1485 0.0622
Final b 2.4631 59.9931 1.9984 1.9849 2.0060 2.0493
parameters c 0.0090 5000.0087 -0.0444 1.0529 -0.0332 0.2732Table 8.15. Classification results of trash objects
in training data using ANFIS: 222-partition
Total number Actual Classified trash type
of objects trash type Bark Stick Leaf Pepper
18 Bark 16 1 1 0
34 Stick 0 32 2 0
54 Leaf 0 1 53 0
107 Pepper 0 0 1 106The identification results of the trash objects in 5 test samples with the 222-
partition are illustrated in Table 8.16. The classification results are superior
compared to all other classifiers discussed so far. The classification rates for
leaf and pepper objects are excellent. However, the classification of bark and
stick objects is satisfactory. As previously mentioned, one of the reasons for
the misclassification of bark and stick objects in the test samples is due to
segmentation defects. The stick object in test sample #1 is a single object but
segmented as a stick and pepper object. During classification, both of these
objects are classified as pepper objects. The stick object is partially buried
under the cotton, resulting in discontinuities of the object in the segmented
image.
Basedontheclassification results of the training data, it is seen that the
ANFIS gives the best classification results with excellent classification rates. Al-
though the classification accuracy of some of the trash types is low, with proper
segmentation it is possible to identify the trash types with increased accura-
cies. The results from the ANFIS indicate that the two sets of membership
partitions, namely, 232-partition and 222-partition, classify the trash types at
similar classifications rates. This is illustrated in Table 8.17.