|
--- |
|
tags: |
|
- image-classification |
|
- timm |
|
library_name: timm |
|
license: mit |
|
datasets: |
|
- cm93/eurosat |
|
pipeline_tag: image-classification |
|
--- |
|
## Model Details |
|
|
|
**Model type:** Convolutional Neural Network (CNN) |
|
|
|
**Finetuned from model :** ResNet50 (pre-trained on ImageNet-1k at resolution 224x224) |
|
|
|
### Model Sources |
|
**Repository:** https://github.com/chathumal93/EuroSat-RGB-Classifiers |
|
|
|
## Training Details |
|
|
|
### Training Data |
|
|
|
The dataset comprises JPEG composite chips extracted from Sentinel-2 satellite imagery, representing the Red, Green, and Blue bands. It encompasses 27,000 labeled and geo-referenced images across 10 Land Use and Land Cover (LULC) classes |
|
|
|
### Training Procedure |
|
**Preprocessing:** Standard image preprocessing including resizing, center cropping, normalization, and data augmentation techniques [RandomHorizontalFlip and RandomVerticalFlip] |
|
|
|
### Training Hyperparameters |
|
- **Learning rate:** 3e-5 |
|
- **Batch size:** 64 |
|
- **Optimizer:** AdamW |
|
- **Scheduler:** PolynomialLR |
|
- **Loss:** CrossEntropyLoss |
|
- **Betas**=(0.9, 0.999) |
|
- **Weight_decay**=0.01 |
|
- **Epochs:** 20 |
|
|
|
|
|
## Evaluation |
|
|
|
### Results |
|
| Model | Phase | Avg Loss | Accuracy | |
|
|:----------------------------:|:----------:|:--------:|:---------:| |
|
| resnet50-eurosat | Train | 0.076420 | 97.56% | |
|
| | Validation | 0.054377 | 98.30% | |
|
| | Test | 0.058930 | 98.07% | |
|
|
|
| Model | Accuracy | Precision | Recall | F1 | |
|
|:----------------------------:|:--------:|:------------:|:-----------:|:-----------:| |
|
| resnet50-eurosat | 98.07% | 0.98078 | 0.98074 | 0.98074 | |
