Abstract:
Classifying twenty distinct Spice varieties will be accomplished with the implementation of
the transfer learning (TL) approach. Bengali cuisine is incredibly flavorful and aromatic, and
the taste comes mainly due to the use of spices. Spices are the sole reason why Bengali
cuisine is so popular. However, background studies observe that there is a severe lack of
acceptable datasets and paperwork for popular spices used in the majority of Bangladeshi
cuisine. Spice classification is significant because it helps upcoming generations by enabling
the recognition of spice varieties without any prior knowledge. To resolve the Spice
recognition problem, collecting a significant dataset on the specific number of images per
class which includes 500 images and a total of twenty spice varieties like Clove, Black
cardamom, Black pepper, Cumin, and Nutmeg etc. The dataset will then be preprocessed by
applying techniques such as Image resizing, and noise reduction. The TL approach will be
used to solve the challenge since it is the most efficient in terms of processing energy,
execution speed, and real-time analysis. Out of all the experiments, the proposed model
"MobileNetV2" outperformed all other models in efficiency, achieving an accuracy score of
99.70% in identifying spice categories. The collected dataset will frequently be helpful for
further studies that classify regional spice variations in Bangladesh.
