Pseudo-SAR Image Generation from Optical Images for Accurate Annotation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for creating training data of synthetic aperture radar (SAR) images face challenges in accurately annotating objects, leading to annotation errors and insufficient data quantity, which affects the accuracy of object detection models.
Innovation Solution
An image generation apparatus and method that utilizes optical images to generate pseudo-SAR images through conversion processing, associating them with annotation data to create training data, thereby reducing annotation errors and increasing data quantity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If annotation is performed directly on SAR images, then training data can be created, but annotation errors occur due to the difficulty of accurately identifying objects in SAR images
Solution Approach 1:
The patent introduces optical images as an intermediary medium. Instead of directly annotating difficult SAR images, the system uses optical images (which are easier to annotate) as a reference and generates pseudo-SAR images from them. This intermediary approach allows accurate annotation to be performed on optical images while still creating training data for SAR image detection models.
Solution Approach 2:
The patent creates pseudo-SAR images as copies or simulations of real SAR images. These pseudo-SAR images are generated by converting optical images through a simulation model, allowing the system to replicate SAR image characteristics without actually capturing them. This copying approach enables accurate annotation transfer from optical to SAR domains.
2Reliability
If more training data is collected, then model accuracy improves, but the time and resources required for data collection increase
Solution Approach 1:
The patent generates pseudo-SAR images as synthetic copies of real SAR images. By creating these simulated images from optical images through conversion processing, the system can rapidly produce large quantities of training data without requiring actual SAR image acquisitions, significantly reducing data collection time and resources.
Solution Approach 2:
The patent performs preliminary conversion of optical images to pseudo-SAR images before they are needed for training. This preliminary action allows the system to pre-generate training data in advance, avoiding the time-consuming process of collecting actual SAR images when training is needed.
3Measurement precision
If optical images are converted to pseudo-SAR images, then annotation accuracy improves, but additional processing steps are required
Solution Approach 1:
The patent uses optical images as an intermediary that bridges the gap between easy-to-annotate images and the target SAR images. The conversion process from optical to pseudo-SAR images is performed once during data preparation, after which the same pseudo-SAR images can be used for multiple training purposes, amortizing the processing complexity over multiple uses.
Data Source
AI summary
An image generation apparatus includes: at least one memory configured to store instructions; and at least one processor configured to execute the instructions to: acquire an optical image in which an object is imaged, acquire annotation data regarding the object applied to the optical image, generate a pseudo-synthetic aperture radar (SAR) image that simulates an SAR image from the optical image by conversion processing, and add the pseudo-SAR image to training data in association with annotation data.