The Research of Multi-scale Effect on Remote Sensing Image Object Detection

Abstract

AbstractObject detectors based on convolutional neural network (CNN) have already achieved state-of-the-art performance due to the powerful feature representation capabilities in computer vision tasks including image classification, object detection, and image segmentation. With the rapid development of remote sensing technology, the object detection of high-resolution optical remote sensing images has become an important part of remote sensing. On account of the large sight and complex background of the remote sensing images, along with the large size variation of the objects, the performance of the methods based on CNN is restricted. This paper will focus on the objects that multi-scale features are of the most obvious to analyze and discuss the impact of multi-scale effect on object detection. A new criterion for setting small objects is presented. Through the analysis of evaluation indicators, we conclude that the multi-scale effect decreases the detection accuracy of small objects under the same condition. Furthermore, a method is proposed to improve the accuracy of small objects detection.KeywordsCNNObject detectionRemote sensingMulti-scale