Object Detection with Deep Learning or Machine Learning

Object detection is a computer vision task that involves identifying and locating objects within an image or video. It combines image classification and localization to detect multiple objects and assign labels to them. Deep machine learning techniques, particularly Convolutional Neural Networks (CNNs) and Transformers, are widely used for object detection due to their high accuracy and performance.

Here are the steps in ArcGIS Pro for object detection using a pre-trained model

Object Detection using Deep Learning

Object Detection using Deep Learning

Object Detection using Deep Learning

Object Detection using Deep Learning

Object Detection using Deep Learning

Object Detection using Deep Learning

Object Detection using Deep Learning

Object Detection using Deep Learning

Object Detection using Deep Learning

Set Up ArcGIS Pro

• Ensure you have the Image Analyst activated.

• Please install ArcGIS Deep Learning Framework from https://github.com/Esri/deep-learning-frameworks

Load Input Imagery

• Open your ArcGIS Pro project and add your imagery: - Click Add Data → Select your raster or satellite imagery.

• Ensure the image is preprocessed (e.g., georeferenced and clipped).

Import Pre-trained Model

• Download the pre-trained model (.dlpk) from https://www.esri.com/en-us/arcgis/deep-learning-models or Living Atlas.

Perform Object Detection

• Open Geoprocessing Tools:

  • Go to Analysis Tab → Tools.

• Search for Detect Objects Using Deep Learning in the Geoprocessing pane.

• Fill in the parameters:

  • Input Raster: Select your imagery.

  • Output Detected Objects: Provide a name for the output feature class.

  • Model Definition: Browse and select the .dlpk file from the pre-trained model.

  • Arguments: Add optional parameters like:

    1. `padding`: Default is 0.

    2. `batch_size`: Adjust for performance (default is 4).

    3. `threshold`: Confidence level for detection (e.g., 0.5).

  • Processing Mode: Select GPU for faster processing if supported.

• Click Run to start the object detection process.

Analyze and Visualize Results

• The output will appear as a feature layer with bounding boxes or polygons around detected objects.

• Customize the symbology using the Symbology pane.

• Use the Attribute Table to review details like confidence scores.

Export and Share Results

• Export results as shapefiles or feature classes for further analysis.

• Share outputs via ArcGIS Online or ArcGIS Enterprise as web layers.

Optional Steps for Customization

• 1. Retrain the Model (if needed)

  • Use the Export Training Data for Deep Learning tool to create labeled data.

  • Fine-tune the model using the Train Deep Learning Model tool.

• 
  

• 2. Model Validation

  • Use test imagery to validate the results and refine parameters.

For additional details about our object detection and pixel classification services or help with any part of the workflow, click here! don’t hesitate to reach out to us at:

Email: info@geowgs84.com 

USA (HQ): (720) 702–4849

India: 98260-76466 - Pradeep Shrivastava

Canada: (519) 590 9999

Mexico: 55 5941 3755

UK & Spain: +44 12358 56710

GeoWGS84 Corp 

https://www.geowgs84.com/services/deep-learning-with-geospatial-data