A lightweight feature fusion network for weak and small target detection in remote sensing

Area of Science:

• Computer Vision
• Remote Sensing
• Artificial Intelligence

Background:

• Remote sensing imagery poses challenges for object detection, including wide fields of view, complex backgrounds, and dense small targets.
• Traditional object detection methods are often ineffective in these demanding remote sensing environments.

Purpose of the Study:

• To develop a lightweight and efficient object detection network tailored for remote sensing applications.
• To improve the accuracy and robustness of small target detection in complex remote sensing scenarios.

Main Methods:

• Introduced GSS-YOLO, a lightweight network integrating a Spatial Information Aggregation (SIA) module within a Cross-Stage Partial Network (C3).
• Incorporated Spatial Pyramid Dilated Convolution (SPD-Conv) for enhanced low-resolution input adaptability.
• Embedded a Global Context-Aware Module (GCAM) before the detection head for refined multi-scale feature representation.

Main Results:

• GSS-YOLO demonstrated superior precision, recall, and robustness on USOD, VisDrone2019, and DIOR datasets.
• The network maintained a lightweight architecture while achieving high performance across color and grayscale imagery.
• Ablation studies validated the effectiveness of the proposed modules.

Conclusions:

• GSS-YOLO offers an efficient and robust solution for small target detection in complex remote sensing.
• The integrated SIA, SPD-Conv, and GCAM modules contribute to improved detection accuracy and processing efficiency.
• This approach addresses key limitations of traditional methods in remote sensing object detection.