AI Security for Geoscience and Remote Sensing: Challenges and Future Trends

Yonghao Xu,Tao Bai,Weikang Yu,Shizhen Chang,Peter M. Atkinson,Pedram Ghamisi

DOI: https://doi.org/10.1109/MGRS.2023.3272825

2023-06-22

Abstract:Recent advances in artificial intelligence (AI) have significantly intensified research in the geoscience and remote sensing (RS) field. AI algorithms, especially deep learning-based ones, have been developed and applied widely to RS data analysis. The successful application of AI covers almost all aspects of Earth observation (EO) missions, from low-level vision tasks like super-resolution, denoising and inpainting, to high-level vision tasks like scene classification, object detection and semantic segmentation. While AI techniques enable researchers to observe and understand the Earth more accurately, the vulnerability and uncertainty of AI models deserve further attention, considering that many geoscience and RS tasks are highly safety-critical. This paper reviews the current development of AI security in the geoscience and RS field, covering the following five important aspects: adversarial attack, backdoor attack, federated learning, uncertainty and explainability. Moreover, the potential opportunities and trends are discussed to provide insights for future research. To the best of the authors' knowledge, this paper is the first attempt to provide a systematic review of AI security-related research in the geoscience and RS community. Available code and datasets are also listed in the paper to move this vibrant field of research forward.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is the security and reliability issues of current artificial intelligence technology in Earth Observation (EO) tasks. Specifically, the paper focuses on how to ensure the security of AI models used in the fields of geoscience and remote sensing (RS), especially their performance and security when facing challenges such as Adversarial Attack, Backdoor Attack, Federated Learning, Uncertainty, and Explainability. Since many geoscience and remote sensing tasks are highly safety - critical, the vulnerability and uncertainty of these models need further attention. By reviewing the current development situation, the paper explores future research opportunities and trends, aiming to provide a systematic review of AI - related security research to promote the research progress in this field.

### Main contributions of the paper
1. **Systematic review**: For the first time, it provides a systematic and comprehensive review of AI - related security research for the geoscience and remote sensing community, covering five aspects: adversarial attack, backdoor attack, federated learning, uncertainty, and explainability.
2. **Theoretical and case analysis**: In each aspect, it provides a theoretical introduction and describes several representative works in detail, emphasizing the potential connection of each case with AI security in EO.
3. **Future prospects**: It summarizes the entire review and proposes four possible research directions: secure AI models, data privacy, trustworthy AI models, and interpretable AI models. At the same time, it identifies the potential opportunities and research trends in each direction to stimulate readers' interest in AI security.

### Main research contents
- **Adversarial attack**: It discusses the principles, methods, and defense strategies of adversarial attacks, especially their application in remote sensing data.
- **Backdoor attack**: It explores the method of introducing hidden triggers through data poisoning during the training stage and how to defend against this attack.
- **Federated learning**: It introduces the concept of federated learning and its application in the fields of geoscience and remote sensing, with the focus on protecting data privacy and security.
- **Uncertainty**: It discusses how to estimate the confidence and robustness of machine learning models, especially their application in Earth Observation tasks.
- **Explainability**: It explores how to understand and interpret machine learning models, especially black - box models such as Convolutional Neural Networks (CNN).

### Future research directions
1. **Adversarial attacks and defenses in different tasks**: At present, most research focuses on scene classification tasks. In the future, it should be extended to other tasks, such as object detection and video tracking.
2. **Adversarial attacks in different forms**: Explore other forms of adversarial attacks besides adding perturbations, such as adversarial patches and natural adversarial examples.
3. **Adversarial attacks in different scenarios**: Research adversarial attacks in black - box settings, especially in cases where attackers have limited access to the training model in practical applications.
4. **Physical adversarial examples**: Explore the possibility of designing adversarial examples in the physical world, considering the complexity and unpredictability of the actual imaging environment.
5. **Positive aspects of adversarial examples**: Research the inherent characteristics of deep - learning models revealed by adversarial examples and explore their potential uses in Earth Observation tasks.

Through these studies, the paper hopes to provide valuable references and guidance for AI security research in the fields of geoscience and remote sensing.