Abstract:Small object detection in remote sensing enables the identification and analysis of unapparent but important information, playing a crucial role in various ground monitoring tasks. Due to the small size, the available feature information contained in small objects is very limited, making them more easily buried by the complex background. As one of the research hotspots in remote sensing, although many breakthroughs have been made, there still exist two significant shortcomings for the existing approaches: first, the down-sampling operation commonly used for feature extraction can barely preserve weak features of objects in a tiny size; second, the convolutional neural network methods have limitations in modeling global context to address cluttered backgrounds. To tackle these issues, a high-resolution network with transformer embedding parallel detection (HRTP-Net) is proposed in this paper. A high-resolution feature fusion network (HR-FFN) is designed to solve the first problem by maintaining high spatial resolution features with enhanced semantic information. Furthermore, a Swin-transformer-based mixed attention module (STMA) is proposed to augment the object information in the transformer block by establishing a pixel-level correlation, thereby enabling global background–object modeling, which can address the second shortcoming. Finally, a parallel detection structure for remote sensing is constructed by integrating the attentional outputs of STMA with standard convolutional features. The proposed method effectively mitigates the impact of the intricate background on small objects. The comprehensive experiment results on three representative remote sensing datasets with small objects (MASATI, VEDAI and DOTA datasets) demonstrate that the proposed HRTP-Net achieves a promising and competitive performance.

A Deep Learning Based Objection Detection Method for High Resolution Remote Sensing Image

Method of Building Detection in Optical Remote Sensing Images Based on SegFormer

Robust Real-Time Object Detection Based On Deep Learning For Very High Resolution Remote Sensing Images

An Optimized Faster R-CNN Method Based on DRNet and RoI Align for Building Detection in Remote Sensing Images

High-Resolution Network with Transformer Embedding Parallel Detection for Small Object Detection in Optical Remote Sensing Images

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

Building Extraction from High-Resolution Remote-Sensing Images Based on Deep Learning

Building Detection via Complementary Convolutional Features of Remote Sensing Images.

Multi-scale Fast Detection of Objects in High Resolution Remote Sensing Images

Hough Transform Guided Deep Feature Extraction for Dense Building Detection in Remote Sensing Images

Object Detection Based on Deep Feature for Optical Remote Sensing Images

Multiconstrained Heterogeneous Deep Network for Remote Sensing Rural Building Detection

Feature Decomposition-Optimization-Reorganization Network for Building Change Detection in Remote Sensing Images

HA U-Net: Improved Model for Building Extraction From High Resolution Remote Sensing Imagery

EFP-Net: High-Precision Remote Sensing Image Object Detection

On-Board Multi-Class Geospatial Object Detection Based on Convolutional Neural Network for High Resolution Remote Sensing Images

Multi-Scale Geospatial Object Detection Based on Shallow-Deep Feature Extraction

A Review: Remote Sensing Image Object Detection Algorithm Based on Deep Learning

Effective Building Extraction From High-Resolution Remote Sensing Images With Multitask Driven Deep Neural Network

Remote Sensing Object Detection Based on Receptive Field Expansion Block

A Deep Learning Approach To Detecting Changes In Buildings From Aerial Images