An Approach To Detect Melanoma Skin Cancer Using Fastai CNN Models

Abstract

Handwritten identification has been one of the top study topics for all researchers. Because of the ease and improved comprehension of their own language scalability with technology. For this purpose, we created a model for handwritten compound Bangla character detection from images. As we know in Bangla language, we have almost 320 compound characters. Among them 171 have been listed searching many google sources. Researchers works with about 110 characters out of 171. So we decided to work with rest of the 50 characters in that list which haven’t worked before. The acquisition of datasets is essential to this research since the dataset will determine how accurate the model is. We have our own dataset collection with over 4500 pieces of data. Data was gathered using sheets and then converted to images (JPG format). Images have been handled in a way that we were able to develop the box detection algorithm in Python open CV along with other processing methods. We have used a machine learning approach which consists of multilayer CNN model with some convolution layer, maxpooling layer, dense layer and dropout layer. For our model, the accuracy we received was 88.48%, which is good.