README.md

---
license: apache-2.0
task_categories:
  - object-detection
  - image-classification
tags:
  - military
  - aircraft
  - aerospace
  - yolo
  - defense
  - commercial-aircraft
  - birds
  - drones
size_categories:
  - 10K-100K
---

# Military Aircraft Detection & Classification Dataset  
### 88 Classes with Advanced Background Suppression

## Overview
This dataset is a professionally curated resource for training high-performance **object detection** and **image classification** models such as **YOLOv11**.  
It contains **88 distinct military aircraft classes** and is explicitly designed for **real-world deployment**, where false positives from civilian aircraft, birds, and small drones are common.

To address this, the dataset incorporates a structured **background suppression strategy**, teaching models not only what *to detect*, but also what *to ignore*.

---

## Dataset Summary

- **Total Images**: 26,668  
- **Military Aircraft Classes**: 87  
- **Image Resolution**: 640 × 640 (uniform)  
- **Annotation Format**: YOLO (`.txt`) with normalized coordinates  
- **Primary Use**: Military aircraft detection and classification  

---

## Dataset Split & Background Statistics

| Split | Total Images | Background Images | Background % |
|------|-------------|------------------|--------------|
| Train | 21,342 | 2,508 | 11.75% |
| Validation | 2,641 | 295 | 11.17% |
| Test | 2,645 | 284 | 10.74% |
| **Total** | **26,668** | **3,127** | **~11.7%** |

The dataset maintains a **stratified split** of approximately **80% Train / 10% Validation / 10% Test** across all 87 classes.

---

## Advanced Background Suppression Strategy

To significantly reduce false detections, the dataset includes **3,127 background-only images** with **empty annotations**.  
These images are intentionally selected to represent common real-world confounders in aerial imagery.

### Background Categories

1. **Empty Skies, Clouds & Commercial Aircraft**  
   Negative samples containing:
   - Clear or cloudy skies with no aircraft  
   - **Commercial passenger and cargo aircraft**  
   This trains the model to distinguish civilian airliners from military platforms.

2. **Bird Backgrounds (≈1.5%)**  
   High-resolution bird imagery to prevent *bird-as-aircraft* false positives, particularly at long range or low resolution.

3. **Commercial Drone Backgrounds (≈1.5%)**  
   Civilian and hobbyist UAVs (quadcopters and small drones), enabling the model to differentiate between commercial drones and military-grade UAVs.

All background images use **empty `.txt` label files (0 bytes)** and contain **no bounding boxes**.

---

## Annotation Format

Each image is paired with a corresponding `.txt` file in YOLO format.

### Positive Sample Example

`su57_01.txt`
68 0.475000 0.496875 0.415625 0.859375

**Field Description**
- `68` → Class ID (Su-57)  
- `0.475000` → X-center (47.5% of image width)  
- `0.496875` → Y-center (49.69% of image height)  
- `0.415625` → Bounding box width  
- `0.859375` → Bounding box height  

### Background (Negative) Samples

Background label files are **intentionally empty**:
- `sky_bg_01.txt`
- `commercial_aircraft_bg_01.txt`
- `birds_v1_01.txt`
- `drones_v1_01.txt`

---

## Final Class ID Table (87 Classes)

| ID | Class | ID | Class | ID | Class | ID | Class |
|----|-------|----|-------|----|-------|----|-------|
| 0 | A10 | 22 | CL415 | 44 | JF17 | 66 | Su34 |
| 1 | A400M | 23 | E2 | 45 | JH7 | 67 | Su47 |
| 2 | AG600 | 24 | E7 | 46 | KAAN | 68 | Su57 |
| 3 | AH64 | 25 | EF2000 | 47 | KC135 | 69 | TB001 |
| 4 | AKINCI | 26 | EMB314 | 48 | KF21 | 70 | TB2 |
| 5 | AV8B | 27 | F117 | 49 | KJ600 | 71 | Tejas |
| 6 | An124 | 28 | F14 | 50 | Ka27 | 72 | Tornado |
| 7 | An22 | 29 | F15 | 51 | Ka52 | 73 | Tu160 |
| 8 | An225 | 30 | F16 | 52 | MQ9 | 74 | Tu22M |
| 9 | An72 | 31 | F18 | 53 | Mi24 | 75 | Tu95 |
| 10 | B1 | 32 | F2 | 54 | Mi26 | 76 | U2 |
| 11 | B2 | 33 | F22 | 55 | Mi28 | 77 | UH60 |
| 12 | B52 | 34 | F35 | 56 | Mi8 | 78 | US2 |
| 13 | Be200 | 35 | F4 | 57 | Mig29 | 79 | V22 |
| 14 | C1 | 36 | FCK1 | 58 | Mig31 | 80 | Vulcan |
| 15 | C130 | 37 | H6 | 59 | Mirage2000 | 81 | WZ7 |
| 16 | C17 | 38 | Il76 | 60 | P3 | 82 | X32 |
| 17 | C2 | 39 | J10 | 61 | RQ4 | 83 | XB70 |
| 18 | C390 | 40 | J20 | 62 | Rafale | 84 | Y20 |
| 19 | C5 | 41 | J35 | 63 | SR71 | 85 | YF23 |
| 20 | CH47 | 42 | J36 | 64 | Su24 | 86 | Z10 |
| 21 | CH53 | 43 | JAS39 | 65 | Su25 | 87 | Z19 |

---

## Intended Use Cases
- Military aircraft detection and classification  
- Civilian vs military aircraft discrimination  
- UAV and drone differentiation  
- Long-range aerial surveillance research  
- False-positive suppression benchmarking for YOLO models