Object Detection Using Filled Images Against Adversarial Attacks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Adversarial example attacks significantly reduce the effectiveness of object detection systems, particularly those based on deep learning methods, by causing erroneous classification results.
Innovation Solution
An object detection device that performs object detection on a target image, generates filled images based on selected bounding boxes, and applies two removal processes to eliminate redundant bounding boxes, thereby neutralizing the adversarial attack.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning methods are used for object detection, then detection accuracy is improved, but vulnerability to adversarial example attacks increases
Solution Approach 1:
The patent segments the detection results into multiple components and processes them through separate detection units. The original image is divided into multiple filled images by masking different regions, and each filled image is processed independently to generate detection results that are then integrated. This segmentation approach reduces the impact of adversarial perturbations on the overall detection accuracy.
Solution Approach 2:
The patent introduces filled images as an intermediary representation between the original image and the final detection result. By generating filled images that mask potential adversarial regions and processing these intermediaries through additional detection units, the system creates a mediating layer that filters out adversarial effects while preserving legitimate object detections.
2Reliability
If multiple detection units and filled images are generated, then robustness against adversarial attacks is improved, but computational complexity increases
Solution Approach 1:
The patent performs preliminary actions by generating filled images and selecting target bounding boxes before the main detection process. Adversarial regions are identified and masked in advance to create filled images, and target bounding boxes are selected based on confidence thresholds before further processing. This preliminary processing reduces the computational burden during the main detection phase by pre-filtering problematic regions.
Solution Approach 2:
The patent applies partial action by selectively processing only certain regions of the image through filled images and additional detection units. Instead of processing the entire image through multiple complex stages, the system focuses computational resources on regions identified as potential adversarial areas or regions containing detected objects, thereby reducing overall computational complexity while maintaining robustness.
Data Source
AI summary
A first detection unit (121) performs object detection on a target image (191) to calculate a detection result (192). A processing unit (130) fills in a target bounding box in the target image for each target bounding box to obtain a filled image group (193). A second detection unit (122) performs the object detection on each filled image to obtain a detection result group (194). A first removal unit (141) performs a first removal process on a detection result set to obtain a first result set (195). A second removal unit (142) performs a second removal process on the first result set to obtain a second result set (196) as an object detection result (197).