It depends on the data size, the no of parameters in the layers. There is no specific model which fits to every application. You can see a model which people have used for similar problem and try to use that and see how it works. After that you can start playing with the model to suit for your own application.

Hello,

1) Each layer performs automatic feature extraction from the input image. The first layer recognizes and filters large and easily detectable parts of the image (features), while the subsequent layers are dedicated to finer and finer features as you add further layers. All these layers are then collectively utilize to detect and classify the objects in the underlying image. Where the last layer does the task of classification.

2) For remote sensing application, you should try to look into papers who have already done the job. For example, this one: https://hal.inria.fr/hal-01369906/document

3) You can find the answer about how many layers to choose and how the decide the architecture in this thesis: https://www.cs.toronto.edu/~vmnih/docs/Mnih_Volodymyr_PhD_Thesis.pdf

Good luck!

The early layers are used for extraction of features. Regarding the number of filters, it depends on the complexity of the tasks you are performing. The last few layers of your models are dense / fully connected layers that are used for performing classification tasks.