Although, I have no expertise in your area of interest but I can guide and suggest you to select an innovative topic for your master's theses in your area of interest.
My suggestions are to check the followings:
1. Call of papers for the relevent journals of your field from the followings publishing groups
emerald, Tylor and Francis, science direct
2. Suggestion for further research (normally in abstract section/conclusion section) from the latest articles in your field.
DL actually depends on the application. For example, you want to work with image processing. In that case, CNN is more suitable than other techniques. If you want to work with time-series data, then RNN, LSTM, GRU are there. That's why first you have to fix the interested area then you can choose by yourself. To do that, first, learn the basics of all DL techniques (ANN, reinforcement learning, for supervised learning -> CNN, RNN, for unsupervised learning -> SOM, DBN, RBM). I have attached a paper that used CNN which might give you an idea of how to apply DL.
Article A Blockchain-Based Artificial Intelligence-Empowered Contagi...
Emerging topics are usually found among those proposed in the call for papers of top conferences, such as CVPR. Recently, I see a lot of interest in eXplainable AI, precision agriculture, remote sensing, human pose estimation, etc. Of course, one domain that attracts a lot of interest today is data analytics for COVID-19.
From a methodological point of view, I see groundbreaking results with (visual) transformers and graph (convolutional) neural networks.