Applications Deep Learning in GIS

The first article in which we will know some details such as:

?Definition of Deep Learning and Neural Networks.

?How do we install and prepare Tools inside Arc Pro.

?We will know how to read and learn about the details of the model before we use it.

?When do we use a ready-made model and when do we train a specific model for area.

• First, we must be aware that GIS Track is always getting updates, and is constantly integrated with Tecnology, through which we can achieve integration of systems, and through Deep Learning, we can save time and money because Automation shortens many steps. Currently, the whole world is turning to using AI. In its environment work, Esri has provided a huge set of tools according to the users’ uses, so that they help in achieving the institutions’ goals in fewer steps. If we make Model from Scratch, it is available to use Notebook ArcGIS or through Geoprocessing Tools, and this will be the core of our articles in the coming period through Arc GIS. Pro. We also cannot forget the enterprises and the service available on it, and it is not possible to run any model through it.
• Deep learning is a branch of artificial intelligence that enables the model to learn by processing data in a way inspired by the human brain and based on neural network technology, which consists of layers in the form of in bouts, and there are different types used in neural networks, including CNN and RNN, which are among the branches of artificial intelligence. I used it for image processing, and we use it with Raster processing, extracting objects from it, and creating different classifications.
• There are a set of basic requirements for installation so that we can use Tools within Arc GIS Pro. I previously talked about them separately. I will leave a link post here https://www.dhirubhai.net/posts/mona-younis-b43232191_esri-arcgispro-activity-7162521901100220417-ZsVH?utm_source=share&utm_medium=member_desktop We must install the deep learning libraries according to the available version of Arc GIS Pro.

• When do we use the pretrained model from ArcGIS Living Atlas and how do we decide on this, and when do we do it from scratch? It depends on the case. I mean, if I want to detect trees in a specific area, what will make me a pretrained model, especially since training takes time and effort while I have a ready-made model that I can use uploaded from Azeri on Atlas, but if I want to detect a specific type of tree here, I have to make the model myself, take a sample, and train it on it in order to reach the highest accuracy. I mean, we can say that if the same requirement is available in a ready-made model, I can use it. With my own region, but if it is not available and the area has different details, here is a new model.
• How do I know the details of the Model from within Arc GIS Pro from Image Classification? I will go to Deep Learning Tools and choose Review Deep Learning Models. The Tool will open with me and choose Model and all Info will appear for it.
• From Atlas, we run a search on the Model, and before we download it, we see its description. The Resolution of the satellite image that connects the Model to the required accuracy will be specified. But does this mean that if the Raster is used with a resolution of, say, 10 meters, the Output will not appear? No, we will find a Layer that has geometry, and it works. Detect, but if it was for buildings, the shape of the building would not be correct, and since the data I had was wrong, I would not be able to benefit from it in any application. However, if Detect was, for example, for the location of a Swimming pool or Trees, we will find a good result that can be relied upon, but it is always preferable to use Raster itself. Resolution is required. For Model.
• From the Toolbox we will use the Tool Detect Objects Using Deep Learning
• parameters

input: Raster

output: save layer

Model: Select on after downloading

? non-maximum

• Environment

select Extent

Sell Size:0.3

processor Type: CPU

Run

The time it takes to run the model depends on the RAM and the graphics card in the device, meaning if a graphics card is RTX 30/50 or above, it will most likely not take more than an hour, as well as the size of the data to which the model is applied.

We will continue, God willing, in the next article, and if there is anything that is not clear to you in Comments.