Improving the Detection of Small Oriented Objects in Aerial Images

Chandler Timm C. Doloriel,Rhandley D. Cajote

DOI: https://doi.org/10.1109/WACVW58289.2023.00023

2024-01-12

Abstract:Small oriented objects that represent tiny pixel-area in large-scale aerial images are difficult to detect due to their size and orientation. Existing oriented aerial detectors have shown promising results but are mainly focused on orientation modeling with less regard to the size of the objects. In this work, we proposed a method to accurately detect small oriented objects in aerial images by enhancing the classification and regression tasks of the oriented object detection model. We designed the Attention-Points Network consisting of two losses: Guided-Attention Loss (GALoss) and Box-Points Loss (BPLoss). GALoss uses an instance segmentation mask as ground-truth to learn the attention features needed to improve the detection of small objects. These attention features are then used to predict box points for BPLoss, which determines the points' position relative to the target oriented bounding box. Experimental results show the effectiveness of our Attention-Points Network on a standard oriented aerial dataset with small object instances (DOTA-v1.5) and on a maritime-related dataset (HRSC2016). The code is publicly available.

Computer Vision and Pattern Recognition,Artificial Intelligence

What problem does this paper attempt to address?

### The Problem the Paper Aims to Solve

This paper aims to address the issue of detecting small oriented objects in aerial images. Specifically:

1. **Current Challenges**: Existing oriented aerial detectors, while performing well in orientation modeling, are not effective in handling small objects. Due to the small pixel area and specific orientation of small objects in images, accurate detection is challenging.

2. **Proposed Method**: The authors propose a new method called the Attention-Points Network to improve the detection accuracy of small oriented objects by enhancing the classification and regression tasks of the oriented object detection model. This network includes two types of loss functions:
- **Guided-Attention Loss (GALoss)**: Utilizes instance segmentation masks as ground truth labels to learn attention features, improving the detection of small objects.
- **Box-Points Loss (BPLoss)**: Scores based on the position of predicted box points relative to the target oriented bounding box.

3. **Experimental Validation**: Experimental results show that this method performs excellently on the standard oriented aerial dataset DOTA-v1.5 (which includes small object instances) and the maritime-related dataset HRSC2016. Additionally, the authors conducted ablation experiments to demonstrate the effectiveness and complementarity of the two loss functions.

In summary, this paper primarily addresses the challenge of detecting small oriented objects in aerial images and proposes a novel and effective solution.