metadata
license: apache-2.0
datasets:
- jonathan-roberts1/NWPU-RESISC45
language:
- en
metrics:
- accuracy
base_model:
- google/efficientnet-b0
pipeline_tag: image-classification
library_name: adapter-transformers
tags:
- climate
EfficientNet-B0 for Satellite Image Classification
Model Description
This model uses EfficientNet-B0 for classifying 45 land-use categories from the NWPU-RESISC45 dataset. Transfer learning and fine-tuning techniques were applied.
Dataset
- NWPU-RESISC45: 31,500 images in 45 classes
- Data augmentation applied: rotation, flips, zoom
Note: Dataset not included in repo; download link below. NWPU-RESISC45 Dataset: https://github.com/remote-sensing-datasets/NWPU-RESISC45
Metrics
- Accuracy: ~84.4% on test set
Intended Use
- Land-use classification from satellite imagery
- Educational or research purposes
Limitations
- Only trained on NWPU-RESISC45, may not generalize to other datasets
- Not intended for production-critical tasks
Example Usage
from tensorflow.keras.models import load_model
from tensorflow.keras.preprocessing import image
import numpy as np
model = load_model("model_weights.h5")
img = image.load_img("example.jpg", target_size=(224,224))
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)/255.0
pred = model.predict(x)
pred_class = np.argmax(pred, axis=1)
print(f"Predicted class: {pred_class}")