I have just started my Ph.D. with the topic of Scene Classification of Images. After reading some of the literature, I have an idea that CNNs with transfer learning has more impact in vision applications. Now questions arise is that there are so many decent works available with high classification accuracies over different datasets (like UCmerced, AID, NWPU, PatternNet, SAT4, SAT6, Eurosat, RSI-CB, MIT67, and so on.).
So,
What is the scope of the work in the mentioned domain? What are the problems that need attention? Which application to focus on (Indoor-outdoor, LULC, Remote Sensing Scenes, or any other available in agriculture or healthcare? If yes then the suggestion will be welcomed.)? Which dataset to work on?
Furthermore, What can we do in it to make a novel one? Shall I focus on a particular dataset? If yes then suggest one.
The respected research community, your minor input can make my day.
In my experience that Artificial intelligence is the future of all sector's, that we can find two types of Machine learning: supervised and unsupervised. For you can starting by supervised by using Transfer Learning or Deep Learning.
For Transfer learning has the model pre-trained, we can you implemented or update to your problems for do it the Classification , for example you can find VGG, ResNet ...etc. model.
For Deep Learning we can create our model by using different architecture, for example NN, CNN, RNN....etc.
The better model is when you get high accuracy and lower loss and we don't get you the over fitting.
For what we can you game for get optimized model or creation the new architecture or creation the new system based Artificial intelligence.....etc. this depended to you problematic or your sector.
We can you find different frameworks for implementation, Caff Tehano Tensorflow kras pytorch......in python.
I hope that be Claire for you and helpful @Sagar M Chavda
In my experience that Artificial intelligence is the future of all sector's, that we can find two types of Machine learning: supervised and unsupervised. For you can starting by supervised by using Transfer Learning or Deep Learning.
For Transfer learning has the model pre-trained, we can you implemented or update to your problems for do it the Classification , for example you can find VGG, ResNet ...etc. model.
For Deep Learning we can create our model by using different architecture, for example NN, CNN, RNN....etc.
The better model is when you get high accuracy and lower loss and we don't get you the over fitting.
For what we can you game for get optimized model or creation the new architecture or creation the new system based Artificial intelligence.....etc. this depended to you problematic or your sector.
We can you find different frameworks for implementation, Caff Tehano Tensorflow kras pytorch......in python.
I hope that be Claire for you and helpful @Sagar M Chavda
Just a thought...
I think in vision applications from accuracy/performance point of view, you should read about YOLOV models which i think are most suitable for obtaining attractive accuracies /performance in computer vision /detection tasks...
Have a look at https://arxiv.org/pdf/1707.01202.pdf
Preprint Remote Sensing Image Scene Classification Meets Deep Learnin...
Article A Survey on Deep Learning-Driven Remote Sensing Image Scene ...
Article A survey of remote sensing image classification based on CNNs
The basic idea behind any research is improvement over existing work. The existing works can be found using a literature survey of good journals in a specific area, which can be considered 'state-of-the-art'. To improve upon them, investigate their limitations and short-comings and propose ways to overcome them. Although scene classification has a lot of scopes in computer vision and robotics, whether it is suitable for you would depend on whether you are able to identify research scopes and feel confident of addressing them.
Besides other sources, you could visit Kaggle. This community can provide a possibility to investigate different solutions from different teams for a lot of ML challenges. You can compare various practical solutions for the same problem in one place.
- Badges
- Science topic