Enhanced Single Shot Small Object Detector for Aerial Imagery Using Super-Resolution, Feature Fusion and...

Area of Science:

• Computer Vision
• Artificial Intelligence
• Remote Sensing

Background:

• Small object detection in aerial imagery is challenging due to high altitudes and wide-angle lenses.
• General object detectors struggle with small objects, losing spatial features and feature representation.
• Imbalance between small objects and background further complicates detection.

Purpose of the Study:

• To address the limitations of current object detectors for small objects in aerial images.
• To propose a novel Convolutional Neural Network (CNN) model for enhanced small object detection.
• To improve feature representation of small objects at the prediction layer.

Main Methods:

• Utilized the Single Shot Multi-box Detector (SSD) as the baseline network.
• Integrated feature enhancement modules: super-resolution, deconvolution, and feature fusion.
• Evaluated the model on three datasets, including two aerial image datasets with predominantly small objects.

Main Results:

• The proposed CNN model demonstrated improved mean Average Precision (mAP) and Recall.
• Achieved superior performance compared to state-of-the-art small object detectors.
• Effectively enhanced feature representation for small objects.

Conclusions:

• The developed CNN model offers a significant advancement in small object detection for aerial imagery.
• The feature enhancement modules are crucial for improving detection accuracy of small objects.
• The model shows promise for applications requiring precise identification of small objects in remote sensing data.