Application of local fully Convolutional Neural Network combined with YOLO v5 algorithm in small target detection of remote sensing image

Area of Science:

• Remote Sensing
• Computer Vision
• Artificial Intelligence

Background:

• Accurate detection of small targets in remote sensing images is crucial for various applications.
• Existing methods often struggle with the scale variation and density of small objects.

Purpose of the Study:

• To develop and evaluate an improved deep learning model for detecting small targets in remote sensing imagery.
• To enhance the detection capabilities of YOLO-v5 by integrating features from Region-Based Fully Convolutional Networks (R-FCN).

Main Methods:

• Comparative analysis of Region-Convolutional Neural Network (R-CNN), Fast Region-Convolutional Neural Network (FRCN), and R-FCN for feature extraction.
• Adaptation of YOLO-v5 algorithm, incorporating a multi-scale anchor mechanism from Faster R-CNN.
• Evaluation of the proposed YOLO-v5 + R-FCN method on NWPU VHR-10 and Vaihingen datasets.

Main Results:

• The YOLO-v5 + R-FCN method demonstrated optimal detection ability, outperforming other algorithms, particularly for small targets like vehicles and storage tanks.
• Achieved high recall rates for diverse small target types and improved accuracy for densely arranged objects.
• The deeper network architecture enhanced feature extraction and recognition capabilities for remote sensing images.

Conclusions:

• The YOLO-v5 + R-FCN approach offers a robust solution for small target detection in remote sensing.
• This research provides valuable insights for applying remote sensing technology and satellite-based target detection.