Ground-based Detection of Martian Dust Devils With a Fine-tuned Fast R-CNN
Description
We developed a two-stage pipeline for efficient dust devil detection in Mars rover imagery. Our approach combines preprocessing filters to remove unsuitable images, followed by a Faster R-CNN with ResNet-50 backbone and Feature Pyramid Network, effectively detecting nonrigid, low-opacity dust devils that traditional methods frequently miss. We demonstrated clear advantages over generic object detection models.
Detailed example
Detecting nonrigid, low-opacity dust devils
AI / analytics pattern
Computer Vision: AI that processes and interprets visual data (e.g., images and videos).
Automation level / stage
a) Pre-deployment – The use case is in a development or acquisition status.
Expected benefit
Our fine-tuned model significantly outperforms general-purpose architectures, highlighting the critical importance of domain-specific training for specialized atmospheric phenomena detection across multiple rover platforms and mission phases. This work has established a foundation for onboard implementation in future Mars missions. Our model could enable intelligent data prioritization, allowing rovers to retain high-resolution imagery of dust devil activity while applying aggressive compression to less scientifically valuable frames, optimizing limited bandwidth resources.
Controls / human review
ATO: Not reported; PIA: Not published