Abstract:
This research develops an efficient and accurate methodology for classifying
the types of coconut in Sri Lanka. At present, coconut-type identification can
only be done by the best professionals and those who are well-trained in the
particular section. Normal people find it difficult to identify the coconut type
accurately before farming. So, as proposed in this research, a simple
application is developed to identify the type of a coconut. CNNs have an
effective architecture compared to other Deep Learning algorithms since they
can detect patterns with high accuracy. Therefore, in this study, a CNN
architecture has been used to identify the shape as a pattern along with its
color. Also, the convolutional layer reduces the high dimensionality of
coconut images without losing its information. CNN automatically recognizes
significant features of coconut images. The coconut images are grouped into
five according to their type, “Typica”, “Navasi”,” Bodiri”, “Eburnea”, and
“Regia”. Here 80% of the images are used for training while 20% of the
images are used for validation. Each group has 1000 images in the dataset. In
building the CNN sequential model, the layers transform one activation to
another through a differentiable function. Here, three main layers are used to
build CNN architecture: the convolution layer, the non-linearity layer, and the
pooling layer. This proposed research announces that the CNN algorithm
reached 85% accuracy perfectly. The future work of this research is to
implement this methodology to develop a mobile application to scan and
identify all types of coconuts in Sri Lanka.