Automated object detection in high-resolution remote sensing satellite images(HRRSSI) is a proper solution for this task rather than manual detection using professional specialists. However, it is more complex due to the varying size, type, orientation, and complex background of the objects to detect. Utilizing artificial intelligence using deep learning is the state of the art technique to achieve this task. The number of labeled satellite images is limited for training a deep neural network therefore; transfer learning techniques were adopted for this task. This paper proposes a framework for airplane detection based on Convolution Neural network (CNN). Faster Region Based CNN (Faster R-CNN) framework is used to perform automatic airplane detection through transfer learning. Inception v2 is added to the network for feature extraction to enhance detection accuracy. The problem of information reduction of the objects due to the resizing of large size satellite image in test phase has been solved by adding a split layer before the input layer, together with a mosaic layer after detection output layer. Dataset is used to build and test the model is collected from Google Earth. Experimental results prove that the proposed developed model is extremely accurate for satellite images object detection.
A Deep Learning Framework for Automatic Airplane Detection in Remote Sensing Satellite Images
2019-03-01
1102812 byte
Conference paper
Electronic Resource
English
Automatic Vehicle Detection from Satellite Images Using Deep Learning Algorithm
Springer Verlag | 2021
|