Electronic Theses and Dissertation

Mushroom poisoning is one of the phenomena that occur in society due to errors in identifying
mushrooms and eating them. The effects are varied, ranging from mild cases, such as nausea,
vomiting, diarrhea, dizziness, to the more severe case, namely death. As part of efforts to
minimize the occurrence of similar incident, a solution is created through the approach of
artificial intelligence, which is to generete a model that can distinguish toxic mushrooms and
non-toxic one. There are five species of mushroom that is carried out in this research, those are
Amanita caesarea, Cantharellus cibarius, and Volvariella volvacea as non-toxic mushrooms and
Amanita phalloides and Omphalotus olearius as toxic mushrooms. There are 142 total images
used in this research. For training data, there are 100 images, with 20 images for each species.
For validation data, there are 37 images in total, with 7 images for each species, except for
species V. volvacea, which has 9 images. And, as for test data, one image for each species. The
model was created using Convolutional Neural Network (CNN). The final model has an
accuracy rate of 78%. During the training process, data augmentation techniques are used,
which is useful to reproduce the same image, but different from original image, by using some
transformations, such as rotation, horizontal and vertical flip, adding noise, affine transform,
blurring, and center crop. There are 7 images generated from the data augmentation process
plus one original image. The batch size used during training phase is 32. Most prediction error
are caused by V. volvacea, where 9 images used as validation in total, only 3 images are
predicted to be correct, the rest are predicted to be A. phalloides. Overall, the model’s
performance is quite good in classifying the species of A. caesarea, A. phalloides, C. cibarius,
and O. olearius but biased against V. volvacea. Therefore, the model produced in this research
is not reliable enough to be applied to the wider community as a “tool” to distinguish these
mushrooms.
Keywords: A. caesarea, A. phalloides, C. cibarius, O. olearius, V. volvacea, Deep Learning,
CNN

Tidak Tersedia Deskripsi