{"id":31642230,"url":"https://github.com/mateluky/xview-satellite-classification","last_synced_at":"2026-04-13T09:31:44.810Z","repository":{"id":317851538,"uuid":"1069086813","full_name":"mateluky/xview-satellite-classification","owner":"mateluky","description":"Classifying satellite imagery objects using deep learning (CNNs, transfer learning) on the xView dataset.","archived":false,"fork":false,"pushed_at":"2025-10-03T11:49:30.000Z","size":2137,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2025-10-23T20:53:24.733Z","etag":null,"topics":["cnn","deep-learning","image-classification","keras","pytorch","satellite-imagery","transfer-learning","xview"],"latest_commit_sha":null,"homepage":"","language":"Jupyter Notebook","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/mateluky.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2025-10-03T11:36:21.000Z","updated_at":"2025-10-03T14:01:16.000Z","dependencies_parsed_at":"2025-10-03T13:32:15.497Z","dependency_job_id":null,"html_url":"https://github.com/mateluky/xview-satellite-classification","commit_stats":null,"previous_names":["mateluky/xview-satellite-classification"],"tags_count":1,"template":false,"template_full_name":null,"purl":"pkg:github/mateluky/xview-satellite-classification","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/mateluky%2Fxview-satellite-classification","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/mateluky%2Fxview-satellite-classification/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/mateluky%2Fxview-satellite-classification/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/mateluky%2Fxview-satellite-classification/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/mateluky","download_url":"https://codeload.github.com/mateluky/xview-satellite-classification/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/mateluky%2Fxview-satellite-classification/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":31746294,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-04-13T09:16:15.125Z","status":"ssl_error","status_checked_at":"2026-04-13T09:16:05.023Z","response_time":93,"last_error":"SSL_connect returned=1 errno=0 peeraddr=140.82.121.6:443 state=error: unexpected eof while reading","robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":false,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["cnn","deep-learning","image-classification","keras","pytorch","satellite-imagery","transfer-learning","xview"],"created_at":"2025-10-07T03:57:18.750Z","updated_at":"2026-04-13T09:31:44.780Z","avatar_url":"https://github.com/mateluky.png","language":"Jupyter Notebook","funding_links":[],"categories":[],"sub_categories":[],"readme":"# xView Satellite Imagery Classification  \n\n*(Deep Learning Course Project – Universidad Politécnica de Madrid, Master in Digital Innovation – EIT Digital)*  \n\n**Authors:**  \n- Ádám Földvári  \n- Joseph Tartivel\n- Máté Lukács\n\n**Instructor:** Roberto Valle  \n\n\n## 📖 Overview  \nThis repository contains the final project for the Deep Learning course at UPM. The objective was to classify objects in high-resolution satellite imagery using progressively advanced deep learning techniques:  \n\n- Feedforward Neural Networks (FFNN)  \n- Regularized FFNNs  \n- Convolutional Neural Networks (CNNs)  \n- Transfer Learning with ResNet50  \n\n\n## 📊 Dataset  \n- **Source:** [xView Dataset](https://xviewdataset.org/)  \n- **Type:** High-resolution satellite images (0.3m GSD, WorldView-3)  \n- **Split:** 761 training images, 85 test images  \n- **Processed:** 21,377 training objects and 2,635 test objects (cropped \u0026 resized to 224×224)  \n- **Classes:** 12 categories (e.g., building, small car, cargo plane, helicopter)  \n\n\n## 🧪 Evaluation Platform  \nFinal testing and evaluation were conducted via a private competition on [Codabench](https://www.codabench.org/). Submissions were provided in the required JSON format and benchmarked against a hidden test set.  \n\n\n## 📂 Project Structure  \n```\n.\n├── ffnn.ipynb # Feedforward Neural Network experiments\n├── reg.ipynb # Regularization strategies for FFNNs\n├── cnn.ipynb # Custom Convolutional Neural Networks\n├── tl.ipynb # Transfer Learning with ResNet50\n├── DeepLearning_JT_AF_ML_finalReport.pdf # Full technical report\n└── README.md # Project overview and methodology\n```\n\nEach notebook corresponds to a development phase, with models iteratively refined at each stage.  \n\n\n## 🔎 Methodology \u0026 Results  \n\n### Phase 1 – Feedforward Neural Networks (FFNN)  \n- Compared shallow vs. deep architectures using flattened image inputs.  \n- Observed overfitting in deeper models due to loss of spatial structure.  \n- **Best test accuracy:** 45.2%  \n\n![FFNN curves](https://github.com/user-attachments/assets/870bd546-5db4-467d-9c51-da25a9e828c1)  \n\n\n### Phase 2 – Regularized FFNNs  \n- Applied batch normalization, dropout, and extended training.  \n- Improved generalization significantly.  \n- **Best test accuracy:** 55.18%  \n\n![Regularized FFNN curves](https://github.com/user-attachments/assets/5d54dbf0-b1ae-4768-a59c-f2ba745791fd)  \n\n\n### Phase 3 – Convolutional Neural Networks (CNNs)  \n- Developed and refined five CNN architectures.  \n- Integrated data augmentation, L2 regularization, batch normalization, dropout, and custom LR schedules.  \n- **Best test accuracy:** 76.36%  \n\n![CNN curves](https://github.com/user-attachments/assets/cf67a9a5-157d-4341-8d94-54f43ff2aaff)  \n\n\n### Phase 4 – Transfer Learning (ResNet50)  \n- Employed a two-stage strategy:  \n  - **Feature extraction** with frozen layers.  \n  - **Selective fine-tuning** of top layers with lower LR.  \n- Achieved highest overall performance with reduced training time.  \n- **Best test accuracy:** 77.87%  \n\n![TL curves](https://github.com/user-attachments/assets/38878945-2439-4ca6-89b0-88fb2f682aca)  \n\n\n## 📈 Results Summary  \n\n| Model                     | Test Accuracy | Precision | Recall |\n|----------------------------|:-------------:|:---------:|:------:|\n| FFNN (Simple)             | 45.2%         | 30.31%    | 33.66% |\n| FFNN + Regularization     | 55.18%        | 44.95%    | 55.50% |\n| Custom CNN                | 76.36%        | 74.00%    | 74.53% |\n| Transfer Learning (ResNet50) | **77.87%** | 67.15%    | 77.47% |\n\n\n## 💡 Key Insights  \n- **Spatially-aware architectures** (CNNs, ResNet50) are critical for image classification tasks.  \n- **Regularization** substantially improves generalization for non-convolutional models.  \n- **Class imbalance** remains a challenge, especially for minority categories (e.g., helicopters).  \n- **Transfer learning** offered the best trade-off between accuracy, recall, and development time.  \n\n\n## ⚙️ Infrastructure \u0026 Evaluation  \n- **Platform:** [Kaggle](https://www.kaggle.com/) with P100 GPU acceleration  \n- **Evaluation:** Codabench private competition with hidden test set  \n- **Submission format:** JSON files for leaderboard evaluation  \n\n\n## 📄 Documentation  \nFor complete methodology, experiments, and analysis, see the [Final Report](DeepLearning_JT_AF_ML_finalReport.pdf).  \n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fmateluky%2Fxview-satellite-classification","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fmateluky%2Fxview-satellite-classification","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fmateluky%2Fxview-satellite-classification/lists"}