Ahnuf/Military_Aircraft_Classification_Dataset

Military Aircraft Classification Dataset

Overview

This dataset is designed for fine-grained object detection of military aircraft and encompasses 96 different military aircraft types. Some types are merged as one class along with their variants because their airframes or external features differ only slightly, making them difficult to distinguish—especially when only parts of the aircraft are visible.

Supported Aircraft Classes

A-10, A-400M, AG-600, AH-64, AKINCI, AV-8B, An-124, An-22, An-225, An-72, B-1, B-2, B-21, B-52, Be-200, C-1, C-130, C-17, C-2, C-390, C-5, CH-47, CH-53, CL-415, E-2, E-7, EF-2000, EMB-314, F-117, F-14, F-15, F-16, F-2, F-22, F-35, F-4, F/A-18, F-CK-1, H-6, Il-76, J-10, J-20, J-35, J-36, J-50, JAS-39, JF-17, JH-7, KAAN, KC-135, KF-21, KJ-600, Ka-27, Ka-52, MQ-25, MQ-9, Mi-24, Mi-26, Mi-28, Mi-8, Mig-29, Mig-31, Mirage2000, P-3, RQ-4, Rafale, SR-71, Su-24, Su-25, Su-34, Su-47, Su-57, TB-001, TB-2, Tejas, Tornado, Tu-160, Tu-22M, Tu-95, U-2, UH-60, US-2, V-22, V-280, Vulcan, WZ-10, WZ-7, WZ-9, X-29, X-32, XB-70, XQ-58, Y-20, YF-23, Z-10, Z-19.

Dataset Structure

The primary data is provided as a single compressed archive for efficient downloading:

• datasetz.zip: Contains all images and annotation files.

Data Partitioning (80-20 Split)

The dataset is divided into train and test folders. To ensure that both folders maintain a consistent 80-20 split for every single aircraft model, we used a technique called Stratified Splitting.

• Train Set (80%): Used for model training and feature learning.
• Test Set (20%): Used for unbiased evaluation of the final model.

This approach ensures that even rare aircraft types are represented proportionally in both the training and testing phases.

Content

The archive contains JPEG images (.jpg) and corresponding annotation files in CSV format with matching filenames. Each aircrafts image with thier information.

Annotations

Each annotation CSV file details the objects within an image using the PASCAL VOC format. The columns include:

• filename: The identifier for both the image and its corresponding annotation file.
• width and height: The dimensions of the image in pixels.
• class: The aircraft type.
• xmin, ymin, xmax, ymax: The coordinates of the bounding box.

Sample Annotation

filename	width	height	class	xmin	ymin	xmax	ymax
000aa01b25574f28b654718db0700f72	2048	1365	F35	852	177	1998	503
000aa01b25574f28b654718db0700f72	2048	1365	JAS39	169	769	549	893
000aa01b25574f28b654718db0700f72	2048	1365	JAS39	125	908	440	1009
000aa01b25574f28b654718db0700f72	2048	1365	B52	277	901	1288	1177