A Feature-Reflowing Pyramid Network for Object Detection in Remote Sensing Images is called FRPNet.

PROJECT TITLE :

FRPNet: A Feature-Reflowing Pyramid Network for Object Detection of Remote Sensing Images

ABSTRACT:

Object detection, which is a significant and fundamental task in the field of remote sensing, has received an increasing amount of attention and research studies in recent years. Nevertheless, geospatial object detection is still difficult because of the dramatic variation in object scales, as well as the intraclass differences and interclass similarities that arise from multiscale and multiclass objects. In this letter, a solution to these issues is suggested in the form of an end-to-end feature-reflowing pyramid network, also known as a FRPNet. There are two benefits of FRPNet that contribute to an improvement in the accuracy of object detection. In the first step of this process, a nonlocal block is inserted into the backbone so that we can determine the relevancy of various regions of the geospatial image when it comes to obtaining discriminative features. In addition, a feature-reflowing pyramid structure is proposed in order to generate a high-quality feature presentation for each scale by fusing fine-grained features from the adjacent lower level. This helps improve the detection capability for multiscale and multiclass objects. Experiments conducted on a publicly available remote sensing data set known as DIOR show that FRPNet is capable of significantly improving performance when measured against a number of state-of-the-art detection methods in terms of mean average precision (mAP).

Did you like this research project?

To get this research project Guidelines, Training and Code... Click Here