Abstract:Multi-class geospatial object detection from high spatial resolution (HSR) remote sensing imagery is attracting increasing attention in a wide range of object-related civil and engineering applications. However, the distribution of objects in HSR remote sensing imagery is location-variable and complicated, and how to accurately detect the objects in HSR remote sensing imagery is a critical problem. Due to the powerful feature extraction and representation capability of deep learning, the deep learning based region proposal generation and object detection integrated framework has greatly promoted the performance of multi-class geospatial object detection for HSR remote sensing imagery. However, due to the translation caused by the convolution operation in the convolutional neural network (CNN), although the performance of the classification stage is seldom influenced, the localization accuracies of the predicted bounding boxes in the detection stage are easily influenced. The dilemma between translation-invariance in the classification stage and translation-variance in the object detection stage has not been addressed for HSR remote sensing imagery, and causes position accuracy problems for multi-class geospatial object detection with region proposal generation and object detection. In order to further improve the performance of the region proposal generation and object detection integrated framework for HSR remote sensing imagery object detection, a position-sensitive balancing (PSB) framework is proposed in this paper for multi-class geospatial object detection from HSR remote sensing imagery. The proposed PSB framework takes full advantage of the fully convolutional network (FCN), on the basis of a residual network, and adopts the PSB framework to solve the dilemma between translation-invariance in the classification stage and translation-variance in the object detection stage. In addition, a pre-training mechanism is utilized to accelerate the training procedure and increase the robustness of the proposed algorithm. The proposed algorithm is validated with a publicly available 10-class object detection dataset.

Object Detection in Optical Remote Sensing Images Based on Positive Sample Reweighting and Feature Decoupling.

Guiding Clean Features for Object Detection in Remote Sensing Images

Learning Critical Features for Arbitrary-Oriented Object Detection in Remote-Sensing Optical Images

Recalibrating Features and Regression for Oriented Object Detection.

Decoupled Classification Refinement: Hard False Positive Suppression for Object Detection

FDLR-Net: A feature decoupling and localization refinement network for object detection in remote sensing images

A Self-Supplementary and Revised Network for Remote Sensing Object Detection

High-Performance Object Detection for Optical Remote Sensing Images with Lightweight Convolutional Neural Networks

Few-Shot Object Detection Based on Contrastive Class-Attention Feature Reweighting for Remote Sensing Images

Remote Sensing Object Detection Based on Receptive Field Expansion Block

FAST-Det: Feature Aligned SSD Towards Remote Sensing Detector

CFC-Net: A Critical Feature Capturing Network for Arbitrary-Oriented Object Detection in Remote-Sensing Images

An Efficient And Robust Integrated Geospatial Object Detection Framework For High Spatial Resolution Remote Sensing Imagery

Feature Enhancement Network for Object Detection in Optical Remote Sensing Images

Revisiting RCNN: On Awakening the Classification Power of Faster RCNN

Deep Adaptive Proposal Network for Object Detection in Optical Remote Sensing Images

Small Object Detection in Optical Remote Sensing Images Via Modified Faster R-CNN

Intelligent Interpretation of High-resolution Remote Sensing Images Based on Deep Learning

Improving remote sensing object detection by using feature extraction and rotational equivariant attention

Rotation-Insensitive and Context-Augmented Object Detection in Remote Sensing Images

Multi-class Geospatial Object Detection Based on a Position-Sensitive Balancing Framework for High Spatial Resolution Remote Sensing Imagery