How to detect solar panels using an Inference Overlay

Editor → Current Situation (Ribbon tab) → Overlays (Ribbon bar)

How to detect solar panels based on an Inference Overlay:

First add a Satellite Overlay. This will be used as input for the Inference Overlay.
Add an Inference Overlay
With the Inference Overlay added and selected, click on the wizard button in the Overlay's detail panel
Once the wizard is ready, click next
In the Select Model step, click on the Select button to open the neural network selection panel.
In the selection panel, select the Neural Network named solar_panel.onnx and click next to advance the wizard.
In the Input Tensors step, verify that multiple input tensors are recognized and click next. These should be correctly configured.
In the Input prequels step, select the Input Prequel 'A', and under Overlay, select the Satellite Overlay. Click next.
In the Output Tensors step, verify that output tensors are present, click next.
In the Settings step, verify that the Score Threshold is around 0.2. This will lower the threshold to accept identified solar panels.
In the Settings step, verify that the Mask Threshold is around 0.90. This will raise the threshold to include pixels of identified solar panels.
Verify that the Inference Mode is set to BBOX_DETECTION and click next.
In the Overlays step, select all result overlay types and click finish.

Inference wizard solar neural network.jpg

Select the Foliage Neural Network.
Inference wizard solar input tensors.jpg

Verify the existence of input tensors.
Inference wizard solar input prequel.jpg

Select Satellite Overlay for prequel A.
Inference wizard solar output tensors.jpg

Verify the existence of output tensors.
Inference wizard solar settings.jpg

Default settings should identify solar panels, but can be tweaked.
Add all result overlays.
Solar inference labels.jpg

Labeled features on 0.1m satellite image after calculation.

How-to's