Abstract:
Convolutional Neural Networks and Deep Learning has revolutionized every field since their
inception. Agriculture, like all other fields, has also been reaping fruits of developments in
mentioned fields. Grapes are one of highest profit yielding and most important fruit related to
the juice and wine industry and even dry fruits in form of raisins. The biggest challenge in
harvesting grape fruit till date is to detect its cluster successfully. Grape is available in different
sizes, colors, seed size and shapes which makes its detection, through simple Computer vision,
even harder. Thus, this research addresses this issue by bringing the solution to this problem
by using CNN and Neural Networks. A dataset was gathered from a grape farm which consisted
of multiple different classes, colors and sizes of grape pictures taken in multiple conditions. It
was split in 80/20 format making the larger chunk training dataset while test set consisted of
20% of the data. This dataset was carefully annotated and then fed to a powerful CNN based
architecture called YOLO. YOLO is written in Darknet and is a very powerful architecture
especially for Image detection. The custom dataset was trained on this architecture and multiple
models were created with accuracy ranging from 86%-92%
