I don’t quite understand the implication of this article. The last paragraph suggests that you are well aware that a pure CNN implementation might be simpler and more accurate. So, is this about not using NNs? Apparently not because you do after all. So is this about presenting an approach that didn’t work as well? It might be, but it doesn’t become very clear from your writing style.
I enjoyed the article but fail to see the point you’re trying to bring across.