Abstract:
"Covering the majority of our body parts, the outer
shell-like structure that protects the human body from any
outcoming harms, is perhaps the skin. Being the most exposed
part, it also suffers from different infectious diseases that causes
the inside organs to be vulnerable too. Although it is pretty
common to be affected by several skin diseases, identifying the
disease flawlessly is often seen to be confusing as the diseases
tend to be hard to distinguish between. Applying computer
vision with a decent trained classification model can come in
really useful in such scenarios. Among vastly available
classification models, not every model can perform similarly in
terms of identifying the precise disease category. To solve this
concern, a custom collected dataset has been gathered,
processed according to needs and afterwards, a transfer
learning model known as “MobileNet-v2” has been trained and
tested. The testing accuracy as demonstrated by the model was
83% in terms of both the testing dataset and unseen images. The
study reflects that, if flawless dataset is ensured and the training
parameters are maintained, accurate skin disease detection can
be automated and at the same time it can be lightweight that
reduces resource usages being a light model. The trained model
can also be useful in medical implementation by taking further
improving techniques."
