Radio Signal Annotation Using Linked Spectrum and Spectrogram Views
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies do not adequately address the generation of teacher data for supervised machine learning in signal detection, particularly in radio signal analysis.
Innovation Solution
An information processing apparatus and method that generates spectrum and spectrogram images from radio signals, allowing users to designate specific signal ranges on these images, superimpose bounding boxes, and output teacher data using the designated ranges and specification information as ground truth data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If spectrum images and spectrogram images are displayed separately, then each image can be analyzed independently, but the user must switch between different displays increasing operation time
Solution Approach 1:
The patent combines the spectrum image display and spectrogram image display on the same screen, allowing users to view both representations simultaneously. This merging of previously separate displays enables direct comparison and correlation between frequency-domain (spectrum) and time-frequency-domain (spectrogram) views, improving detection accuracy while reducing the time needed to analyze signals across different representations.
2Manufacturing precision
If manual annotation of signal ranges is performed without automated assistance, then precise ground truth data can be obtained, but the workload and time consumption increase significantly
Solution Approach 1:
The system performs preliminary automated detection to generate initial bounding box annotations on both spectrum and spectrogram images before presenting them to users for verification. This preliminary action provides users with pre-processed annotations that are already close to the correct signal ranges, reducing the manual effort required to achieve precise ground truth data while maintaining high accuracy through user verification of the automated results.
3Ease of operation
If only spectrum images or only spectrogram images are used for annotation, then the annotation process is simpler, but the resulting teacher data lacks comprehensive signal characteristics
Solution Approach 1:
The patent segments the annotation task into two complementary views: spectrum image annotation for frequency-domain characteristics and spectrogram image annotation for time-frequency characteristics. Users perform annotation on one view, and the system automatically generates corresponding annotations on the other view, ensuring that both frequency and temporal signal characteristics are captured in the final teacher data without requiring users to manually annotate both views separately.
Data Source
AI summary
To appropriately generate teacher data for supervised machine learning. Provided is an information processing apparatus including: a generation unit that generates a spectrum image and a spectrogram image based on spectrum data of a received radio signal; a display unit that displays the spectrum image and the spectrogram image on a same screen, accepts designation of a specific range including a specific signal from a user on one of the spectrum image and the spectrogram image, superimposes and displays a first bounding box corresponding to the specific range on the spectrum image, and superimposes and displays a second bounding box corresponding to the specific range on the spectrogram image; and an output unit that outputs teacher data in which information indicating the specific range and specification information of the specific signal are used as ground truth data and the spectrum image is used as an explanatory variable.