dc.description.abstract |
Recent advancement of computer vision has made object detection from images much easier,
however, automatically classifying fruits using computer vision still remains a challenging task due
to similarities between different types and various factors like their position (e-g stacked) or
lighting conditions. A fruit classification system can play a vital role in major fields like
autonomous agricultural robotics or simply be used in developing mobile applications for detecting
specific fruit species on the market. In this paper, we evaluated 5 different models that used deep
convolutional neural network (DCCN) techniques for fruit detection and proposed an efficient
model based on our training results. VGG-16, RESNETV2-152, INCEPTION-V3, EXCEPTION,
DENSENET-201 was used to train with fruits that are endemic to Bangladesh. Our dataset
contained fruit images belonging form 7 different classes of native fruits. 80 percent of the dataset
was utilized for training and the other 20 percent for testing purposes. The training dataset was
augmented and increased for better training convenience. We conducted the experiment with our
own dataset and the VGG-16 model achieved a high accuracy rate of 100%. |
en_US |