Multi U-Net architecture has transformed aerial image object detection, pioneering advancements in geospatial analysis. This modified U-Net model excels in complex multi-class segmentation scenarios, leveraging spatial information effectively. The project employs a sophisticated image processing pipeline, maximizing deep learning model training through Min-Max scaling and Patchify techniques. It utilizes performance evaluation metrics like the Jaccard coefficient and a custom loss function hierarchy combining Focal Loss and Dice Loss for efficient model training. Results demonstrate Multi U-Net's ability to handle specific item types, reduce false positives/negatives, and adapt to diverse datasets and domains. Its improved segmentation precision benefits environmental monitoring, disaster management, and urban planning, showcasing its potential for impactful decision-making processes across various disciplines.
Humans in the Loop has published an open access dataset annotated for a joint project with the Mohammed Bin Rashid Space Center in Dubai, the UAE. The dataset consists of aerial imagery of Dubai obtained by MBRSC satellites and annotated with pixel-wise semantic segmentation in 6 classes.
The dataset includes 72 images grouped into 8 larger tiles. The images are labeled and contain the following classes:
Name | R | G | B | Color |
---|---|---|---|---|
Building | 60 | 16 | 152 | |
Land | 132 | 41 | 246 | |
Road | 110 | 193 | 228 | |
Vegetation | 254 | 221 | 58 | |
Water | 226 | 169 | 41 | |
Unlabeled | 155 | 155 | 155 |
The trained models can be found here.
Predictions on Validation Set Images: