Satellite-Image-Classification
- Built an Image Classification model to predict the type of geographical area that is being represented in a satellite image.
- Compared 4 different CNN architectures ie. a custom CNN, InceptionV2, MobileNetV2 , EfficientNetB3 based on model performance on this dataset.
Data
This dataset has 4 different classes of satellite images taken from google maps. The objective is to predict the images into 4 classes: cloudy , desert , green_area , desert.
The dataset can be downloaded from here
Experiments
MobileNetV2:
- The MobileNetV2 model was initialized with pre-trained ImageNet weights and all the layers were fine-tuned. Training the model with an Adam optimizer with learning rate of 0.001 for 8 epochs yielded an Accuracy of 99.2%.
Custom CNN:
- A baseline was created using a custom CNN model with 2 convolution layers and 3 dense layers. A kernel of size 3 x 3 was used for all the convolution layers. A dropout layer was used with dropout rate 0.1 .Training the model with an Adam optimizer with learning rate of 0.001 for 5 epochs yielded an accuracy of 90%.
InceptionV2:
- The InceptionV2 model was initialized with pre-trained ImageNet weights. Only the Dense layers were fine-tuned. Training the model with an Adam optimizer with learning rate of 0.001 for 5 epochs yielded an Accuracy of 88.92%.
EfficientNetB3:
- The EfficientNetB3 model was initialized with pre-trained ImageNet weights and all the layers were fine-tuned. Training the model with an Adam optimizer with learning rate of 0.001 for 5 epochs yielded an Accuracy of 86.72%
| Model | Accuracy |
|---|---|
| Baseline CNN | 90% |
| Mobilenetv2 | 99% |
| InceptionV2 | 88% |