Data and beyond

In this blog post, it will be discussed the possible data samples for possible final projects. First of all, I hardly was able to decide what to do in master thesis project, till yesterday. I was thinking a few possible projects and did research on these. Before writing anything related main topic of today's blog, I never thought that, choosing topic for final project would be that hard, namely time consuming. I researched about a few different topics but will point out 2 most interesting ones.

First topic is about Plant Disease Detection. I researched, and checked the papers about this problem. There are numerous of them who tried to solve issue but they were more specific in group of plants. For example, there is a competition in Kaggle, in which they tried to identify the category of foliar diseases in apple trees. Here is the sample, data link that is provided in competition page. I checked some of them, and tried to re-run their samples and test. My idea about this problem was more general approach, which is really challenging for one project. For instance, in that competition, they tried to detect if leaf has one, two or more than 2 diseases, however these diseases are known beforehand. Hence, system was learned with the samples of these diseases, so if new one comes out, system won't be able reveal what this one is. My idea was to develop a system which detects all (in the best case, nearest to the all) and if there is any disease on leaf or not. After, analyzing the fact that how many different categories exist in today's plant pathology, I step back and think a while. I guess, it would be extremely challenging to do that, therefore, continued to search.

Finally, one of my groupmates told me his idea and I liked that. Because, I think it is one of the most interesting and useful project among the others I looked for. Thus, we discussed a few concepts of project, and started to collaborate. His idea is detection of automobile license numbers for specific purposes like automatic gates/entrances for vehicles. To train that model, we definitely will need dozens of image/video (mainly video) files. Actually, that's only one part of that project. Finding proper data. As our main focus will be detection vehicle registration plates in Azerbaijan, we will need to find good and reliable source of data for that purpose. This data set is sample from Kaggle. We will use the similar, but our purpose to do that based on stream/video data. Also, there are another aspects in that project which should be implemented after doing that part.

E.O.F