ORSIm Detector: A Novel Object Detection Framework in Optical Remote Sensing Imagery Using Spatial-Frequency Channel Features (1901.07925v2)

Abstract: With the rapid development of spaceborne imaging techniques, object detection in optical remote sensing imagery has drawn much attention in recent decades. While many advanced works have been developed with powerful learning algorithms, the incomplete feature representation still cannot meet the demand for effectively and efficiently handling image deformations, particularly objective scaling and rotation. To this end, we propose a novel object detection framework, called optical remote sensing imagery detector (ORSIm detector), integrating diverse channel features extraction, feature learning, fast image pyramid matching, and boosting strategy. ORSIm detector adopts a novel spatial-frequency channel feature (SFCF) by jointly considering the rotation-invariant channel features constructed in frequency domain and the original spatial channel features (e.g., color channel, gradient magnitude). Subsequently, we refine SFCF using learning-based strategy in order to obtain the high-level or semantically meaningful features. In the test phase, we achieve a fast and coarsely-scaled channel computation by mathematically estimating a scaling factor in the image domain. Extensive experimental results conducted on the two different airborne datasets are performed to demonstrate the superiority and effectiveness in comparison with previous state-of-the-art methods.

• The paper introduces the ORSIm detector that integrates spatial-frequency channel features to create rotation-invariant and robust representations in remote sensing imagery.
• It employs feature learning refinement and a fast image pyramid matching technique to efficiently process multiscale objects with reduced computational cost.
• Experimental results on satellite and aerial datasets show significantly improved detection precision and recall compared to state-of-the-art methods.

A Novel Object Detection Framework for Optical Remote Sensing Imagery

Recent advances in spaceborne imaging technologies have significantly expanded the availability of optical remote sensing imagery, leading to an increased interest in automating object detection tasks using this data. Addressing challenges inherent to remote sensing data, such as image deformations due to scaling and rotation, this paper introduces a novel object detection framework: the Optical Remote Sensing Imagery Detector (ORSIm detector). This framework incorporates a comprehensive feature extraction procedure, learning-based refinement, and an efficient testing process utilizing fast image pyramid techniques.

Main Contributions

The ORSIm detector seeks to enhance object detection in optical remote sensing by solving limitations related to incomplete feature representations found in common methodologies. The framework brings together several innovative components:

1. Spatial-Frequency Channel Features (SFCF): The detector integrates spatial and frequency domain information to address challenges posed by image deformations. The SFCF approach constructs rotation-invariant features in the frequency domain while simultaneously leveraging spatial channel features, including color and gradient magnitude, to form a robust feature representation.
2. Feature Learning and Refinement: After extracting SFCF, a feature learning strategy is employed to derive high-level, semantically meaningful features in a reduced-dimensional space. This step enhances the framework's ability to generalize and effectively discriminate between different object classes under varying conditions.
3. Fast Image Pyramid Matching: During testing, ORSIm employs a fast pyramid generative model to efficiently process multiscaled objects, thus reducing computational complexity without adversely affecting detection performance. This model leverages fractal statistics of natural images to approximate feature extraction for multiple scales swiftly.
4. Boosting Ensemble Classifier: A boosting-based ensemble classifier, AdaBoost, is utilized to train the detector with diverse tasks, providing adaptability and improved accuracy in differentiating between positive and negative instances.

Experimental Verification

The ORSIm detector was rigorously tested using two different optical remote sensing datasets: a satellite dataset focused on vehicle detection and the NWPU VHR-10 dataset, primarily assessing airplane detection. Extensive experiments demonstrated that this framework consistently outperformed prior state-of-the-art techniques, achieving higher detection precision and recall rates. Results indicated remarkable performance robustness to variations in object orientation and scale, vital for practical applications in optical remote sensing environments.

Implications and Future Work

The development of ORSIm introduces several implications for optical remote sensing. Practically, it enhances automatic object detection capability, proving valuable for applications such as environmental monitoring and hazard response. Theoretically, it challenges and expands the understanding of feature representation in remote sensing, especially concerning rotation and scale invariances.

Looking forward, future research may focus on adapting the ORSIm framework for end-to-end deep learning implementations, with potential to incorporate more sophisticated hierarchical feature learning mechanisms. Additionally, the integration of tiny object detection capabilities and multi-target recognition in a unified model could drive further advancements in the domain, broadening the applicability of automated detection systems in complex remote sensing imagery scenarios.