Aircraft Detection and Classification using Satellite Imagery (ADCSI)

Abstract

In the modern era, satellite or drone imagery is easily accessible. There are several uses for such images, including the detection and identification of desired targets like aircraft, convoys, trains, and trucks. It can also be used to identify infrastructure, like runways, storage buildings, Air bases and Airports. Our effort adds the identification and classification of aircraft in Google Earth Imagery as another extremely effective use of overhead imagery, broadening the scope of these applications. This can be very helpful for locating and documenting an aircraft in a particular area. In this study, a large number of multi-resolution satellite images were used to train the Convolutional Neural Network-based machine learning algorithm YOLO V5. By selecting training parameters optimized by learning from multiple literature sources and testing them, the models were trained. After a period of extensive model training and achieving desirable accuracy, two user interfaces were developed. Users can detect in real-time when browsing Google Earth or any other source of overhead imagery with the use of UI Live Detection Mode. In Google Earth's Auto-Scan mode, a predefined area is automatically scanned, and all detections are recorded along with classification information. Along with the pixel values of aircraft in the Google Earth image that was acquired, the UI can precisely provide the specific geographic coordinates of the aircraft.