lostmartian/building-and-road-segmentation-from-aerial-images

Aerial Image segmentation using different EfficientNet based backbone encoders with UNet on Massachusetts Building and Road dataset

Building and Road Segmentation from Aerial Images using EffUNet

In city, information about urban objects such as water supply, railway lines, power lines, buildings, roads, etc., is necessary for city planning. In particular, information about the spread of these objects, locations and capacity is needed for the policymakers to make impactful decisions. This thesis aims to segment the building and roads from the aerial image captured by the satellites and UAVs. Many different architectures have been proposed for the semantic segmentation task and UNet being one of them. In this thesis, we propose a novel architecture based on Google’s newly proposed EfficientNetV2 as an encoder for feature extraction with UNet decoder for constructing the segmentation map. Using this approach we achieved a benchmark score for the Massachusetts Building and Road dataset with an mIOU of 0.8365 and 0.9153 respectively.

Tech Stack

Libraries: PyTorch, Numpy, Matplotlib, SMP

Results

Evaluation of different models for building dataset

Model	mIOU	Dice Loss	Precision	Recall	F1 Score	Accuracy
V2S+UNet	0.8159	0.1054	0.8746	0.9220	0.8977	0.8997
V2M+UNet	0.8293	0.0977	0.8821	0.9316	0.9062	0.9080
B7+UNet	0.8359	0.0934	0.8863	0.9352	0.9101	0.9119
V2L+UNet	0.8365	0.0925	0.8865	0.9356	0.9104	0.9122

Evaluation of different models for road dataset

Model	mIOU	Dice Loss	Precision	Recall	F1 Score	Accuracy
V2S+UNet	0.9139	0.0453	0.9321	0.9786	0.9548	0.9558
V2M+UNet	0.9140	0.0475	0.9323	0.9786	0.9549	0.9559
V2L+UNet	0.9147	0.0468	0.9328	0.9790	0.9553	0.9563
B7+UNet	0.9153	0.0461	0.9332	0.9792	0.9556	0.9566

Building Segmentation

Road Segmentation