A DATASET AT A GLANCE...

Planes in Satellite Imagery Dataset

 Accurate detection of airplane from satellite images, which are more cloudy, snowy and foggy with multiple objects in it, is a very tedious process. Despite the difficulty encountered during detection, its valuable application in the monitoring and management of air traffic between take-off and landing and the identification of sabotaging aircrafts in military applications have necessitated airplane detection from satellite images, either manually or through automatic approaches. However, manual detection is more cumbersome that automation becomes completely essential. Hence, machine learning and computer vision- based approaches have a great scope in airplane detection from satellite imagery. Researchers with deep interest in serving the above-mentioned applications can make use of the “Planes in Satellite Imagery” dataset from the Kaggle repository (https://www.kaggle.com/datasets/rhammell/planesnet). Thought the dataset could be directly downloaded from the same site, the source of this dataset is the Planet's Open California dataset, which encompassed 3-meter orthorectified Planet Scope images being collected over California. Containing about 32000 images as RGB images with a size of 20×20 in PNG format, this dataset includes about 8000 airplane-centred images in varied dimensions, angles and atmospheric conditions for classifying them to be ‘plane’. The remaining 24000 images exclude planes, either completely or partially, allowing the classification result as ‘no plane’, ‘partial plane’ and ‘mislabelled bright pixels or strong line features’. Hence, each distinct image filename in the dataset is provided as: {label} _ {scene id} _ {longitude} _ {latitude}.png, enabling researchers to easily avail them for automation purposes.

Image credit: www.unsplash.com