Automatic Detection of Solar Panels in High-Resolution Aerial Imagery

📖 Contents

• Introduction
• Requirements
• Involved Models
• Dataset
• Trained Models
• Evaluations
• Run Solar Panel Detection
• Demo

🏃 Introduction

The global shift toward renewable energy has led to a rapid increase in the deployment of solar panels across residential, commercial, and industrial areas. Accurately mapping the location and extent of these installations is crucial for energy infrastructure planning, policy-making, and environmental monitoring. With the growing availability of high-resolution aerial imagery and advances in computer vision, automated detection methods offer a scalable and cost-effective alternative. This project aims to develop a system that automatically detects solar panels in aerial images using state-of-the-art computer vision techniques.

⚡ Requirements

tensorflow = 2.16.2
opencv = 4.11.0
matplotlib = 3.10.1

🍞 Involved Models

• SegNet
• Fast SCNN

🐶 Dataset

• Dataset are made with high-resolution aerial imagery from Nearmap
• Labelme is used as the tool to label the images.
• Training dataset contains 3936 256x256 rgb images and labels which are collected from Capalaba, Springfield, New Farm, Fairfield, Sunnybank Hills in Brisbane, Australia.
• Validation dataset contains 1344 images and labels which are collected from Springfield and Sunnybank Hills.
• Test dataset contains 1360 images and labels which are collected from a suburb in Perth, Australia.
• Dataset will be made available in the near future.

🐱 Trained Models

• SegNet 0: SegNet with 5 encoders and 5 decoders (Original SegNet).
• SegNet 1: SegNet with 4 encoders and 4 decoders.
• SegNet 2: SegNet with 5 encoders and 5 decoders, each encoder is replaced by a ResNet block (Block with 3 convolutional layers).
• SegNet 3: SegNet with 5 encoders and 5 decoders, each encoder is replaced by a ResNet block (Block with dynamic number of convolutional layers).
• Fast SCNN 0: Original Fast SCNN.
• Fast SCNN 1: Fast SCNN with the first two DSConv layers removed, modified Upsample layers.
• Fast SCNN 2: Fast SCNN with the first two DSConv layers replaced with Conv layers, modified Upsample layers.

🐨 Evaluations

Model Name	Evaluation IoU
SegNet 0	0.8047909
SegNet 1	0.8196788
SegNet 2	0.78407985
SegNet 3	0.7865806
Fast SCNN 0	0.6196553
Fast SCNN 1	0.72767824
Fast SCNN 2	0.82243156

Run Solar Panel Detection

• Create "test_images" under the root project folder. Prepare high-resolution aerial/satellite images with arbitrary size. Place these images inside "test_images" folder. A test image example can be downloaded here.
• Create "prediction_images" under the root project folder. The prediction of the model will be saved here.
• Modify "solar_panel_detection.py". In line 16, modify "IMAGE_LIST", put test image names inside.
• Modify "solar_panel_detection.py". In line 118, modify "model_type" (segmentation model), "model_name" (model weights name), "test_image_name", and "saving_image_name" .
• Run the following command to perform solar panel detection on the given test image.

python solar_panel_detection.py 

Demo

Demo 1

Demo 2