🎉 The research on Planetary Image Classification has been successfully completed

🎉 Welcome 👋

Planetary Image Classification with Transfer Learning 🪐

Built an end-to-end pipeline with PyTorch + torchvision backbones (EfficientNet-B3 / ResNet50): data prep & augmentation → two-stage transfer learning → evaluation & interpretability (lightweight Grad-CAM) → batch/single inference with visual & CSV outputs. We have achieved 100% accurate recognition of Earth, Jupiter, Mars, Moon, Neptune, Saturn, and Uranus.

Research Background 📌

Automatically classify planetary images into 7 classes (default: earth, jupiter, mars, moon, neptune, saturn, uranus). The pipeline is production-minded, reproducible, interpretable, and supports batch inference with visual reporting.

Research Methodology 🔍

1️⃣Frameworks: PyTorch, torchvision, scikit-learn, NumPy, Pandas, Matplotlib/Seaborn, OpenCV, TensorBoard

2️⃣Models: Transfer learning (EfficientNet-B3 / ResNet50 pretrained), custom FC head (ReLU + Dropout)

3️⃣Training: Two stages (freeze features → train head; unfreeze tail → fine-tune), cosine annealing + warmup, class weights, strong augmentations

4️⃣Evaluation: Classification report, confusion matrices (raw/normalized), per-class accuracy, prediction grids, lightweight Grad-CAM

5️⃣Inference: Batch prediction with visual outputs and CSV summary; showcase collage for blind-test mapping

Environment & Dependencies 💻

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

pip install scikit-learn numpy pandas matplotlib seaborn pillow tqdm opencv-python tensorboard

requirements

torch
torchvision
torchaudio
scikit-learn
numpy
pandas
matplotlib
seaborn
pillow
tqdm
opencv-python
tensorboard

⚠️ The project has been uploaded to GitHub, You can access the source code and more details through this 👉PlanetaryImageClassifier 🦄✨