|
|
--- |
|
|
language: |
|
|
- en |
|
|
license: mit |
|
|
library_name: pytorch |
|
|
tags: |
|
|
- computer-vision |
|
|
- image-classification |
|
|
- deep-learning |
|
|
- convnext |
|
|
- pytorch |
|
|
- forestry |
|
|
- plants |
|
|
- environment |
|
|
- ai-for-good |
|
|
pipeline_tag: image-classification |
|
|
model_name: FloraGuard |
|
|
model_type: ConvNeXt-Base |
|
|
base_model: timm/convnext_base |
|
|
author: Aarav |
|
|
description: > |
|
|
FloraGuard is a deep learning–based image classification model designed to |
|
|
identify and classify forest and plant-related categories. It uses a |
|
|
ConvNeXt-Base backbone pretrained on ImageNet and a custom classification head |
|
|
for improved accuracy and generalization. |
|
|
training: |
|
|
framework: PyTorch |
|
|
optimizer: AdamW |
|
|
loss_function: CrossEntropyLoss |
|
|
batch_size: 4 |
|
|
epochs: 20 |
|
|
scheduler: ReduceLROnPlateau |
|
|
device: CUDA / CPU |
|
|
architecture: |
|
|
backbone: ConvNeXt-Base (pretrained on ImageNet) |
|
|
classification_head: | |
|
|
Linear → BatchNorm → ReLU → Dropout → Linear |
|
|
dataset: |
|
|
format: ImageFolder |
|
|
structure: | |
|
|
Dataset/ |
|
|
├── train/ |
|
|
│ ├── class_1/ |
|
|
│ ├── class_2/ |
|
|
│ └── ... |
|
|
└── val/ |
|
|
├── class_1/ |
|
|
├── class_2/ |
|
|
└── ... |
|
|
description: | |
|
|
Multi-class forest and plant image dataset organized by class folders. |
|
|
augmentation: |
|
|
train: |
|
|
- Resize (320x320) |
|
|
- Random Horizontal Flip |
|
|
- Random Rotation |
|
|
- Color Jitter |
|
|
- Normalization |
|
|
validation: |
|
|
- Resize (320x320) |
|
|
- Normalization |
|
|
evaluation: |
|
|
metrics: |
|
|
- F1 Score |
|
|
- Confusion Matrix |
|
|
- Classification Report |
|
|
- Training & Validation Loss Curves |
|
|
outputs: |
|
|
best_model: best_model.pth |
|
|
use_cases: |
|
|
- Forest monitoring |
|
|
- Plant classification |
|
|
- Environmental AI applications |
|
|
- Research and education |
|
|
requirements: |
|
|
- torch |
|
|
- torchvision |
|
|
- timm |
|
|
- numpy |
|
|
- matplotlib |
|
|
- scikit-learn |
|
|
- pillow |
|
|
datasets: |
|
|
- yunusserhat/FireRisk_Original |
|
|
--- |
|
|
|
|
|
|
|
|
**Evaluation Results** |
|
|
|
|
|
*Classification Report & Confusion Matrix------>* |
|
|
|
|
|
 |
|
|
|
|
|
|
|
|
**Notes** |
|
|
|
|
|
-*Overall accuracy:* 0.63 or 63% |
|
|
|
|
|
-*Best performing classes:* |
|
|
|
|
|
1.Non-burnable (F1: 0.85) |
|
|
|
|
|
2.Very_Low (F1: 0.74) |
|
|
|
|
|
3.Water (F1: 0.78) |
|
|
|
|
|
-*Weaker performance on:* |
|
|
|
|
|
1.Moderate |
|
|
|
|
|
2.High |
|
|
|
|
|
3.Low |
