Time-Series Image Patch Attack Detection With Localized Masking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional motion analysis models are vulnerable to patch-based adversarial attacks, which can disrupt decision-making capabilities in applications like self-driving vehicles and camera systems, leading to potential safety hazards.
Innovation Solution
A machine learning model is trained to detect patch-based adversarial attacks in time-series image data, identifying anomalies in both space and time, and modifies the image data to mitigate these attacks, ensuring the integrity of image processing models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a machine learning model processes time-series images to detect adversarial patches, then the detection accuracy improves, but the processing time and computational complexity increase
Solution Approach 1:
The system segments the image processing task by dividing it into multiple processing stages: initial anomaly detection, locality identification, and selective modification. This segmentation allows the system to process only relevant portions of images at high detail, reducing overall processing time while maintaining detection accuracy.
Solution Approach 2:
The system applies partial action by performing full anomaly detection only on regions identified as containing anomalies, rather than processing entire images at full resolution. This selective processing significantly reduces computational burden while maintaining detection effectiveness.
2Reliability
If the system modifies image data to mask anomalies, then the reliability of image processing improves, but the complexity of the system increases
Solution Approach 1:
The system introduces an intermediary component that acts as a bridge between anomaly detection and the image processing model. This intermediary selectively modifies only those image regions containing anomalies, rather than requiring complete system redesign, thus improving reliability with minimal added complexity.
Solution Approach 2:
The modification approach applies local quality by targeting only specific regions of images containing anomalies for modification, rather than applying uniform processing across entire images. This localized approach improves reliability where needed while minimizing overall system complexity.
3Measurement precision
If the system processes all time-series images at full resolution, then the detection capability improves, but the computational resources required increase
Solution Approach 1:
The system performs full-resolution processing only on image regions containing anomalies rather than processing entire images at full resolution. This partial processing approach maintains detection capability for anomalies while significantly reducing computational resource consumption on normal image regions.
Solution Approach 2:
The system segments image processing into different resolution levels, processing only anomaly-containing regions at full resolution while using lower resolution for other areas. This segmentation strategy preserves detection capability where needed while reducing overall computational resource requirements.
Data Source
AI summary
A present invention embodiment prevents patch-based adversarial attacks. A plurality of time-series images are processed using a machine learning model to identify an anomaly, wherein the anomaly comprises an adversarial patch-based attack, and wherein the anomaly is present in a locality comprising a particular time and space in the plurality of time-series images. A subset of the plurality of time-series images are modified based on the locality of the anomaly to mask the anomaly from detection by a trained image processing model.