Spectrogram Color Mapping for Accurate Deep Learning Signal Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods fail to accurately detect or identify signals from spectrogram data using deep learning due to differences in degrees of freedom between RGB and HSV color spaces, and complex number values in time-frequency planes, leading to inappropriate processing.
Innovation Solution
A signal processing device employing window function, Fourier transform, color space conversion, and deep learning processing to convert phase angles of complex amplitudes into hues and associate absolute values with brightness, using a two-dimensional plane of time and frequency for accurate signal detection and identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If conventional color space conversion methods (RGB to HSV) are applied to spectrogram data, then the processing follows standard color space conversion procedures, but the detection and identification accuracy of signals deteriorates due to mismatched degrees of freedom between complex numbers and color values
Solution Approach 1:
The patent transforms the parameter representation by mapping the real part of complex numbers to the L (lightness) channel and the imaginary part to the A (green-red opponent) and B (blue-yellow opponent) channels of LAB color space. This parameter transformation resolves the degree of freedom mismatch while preserving all information from the original spectrogram data.
Solution Approach 2:
The patent introduces LAB color space as an intermediary representation between the complex number domain and the deep learning processing stage. This intermediary space maintains the two degrees of freedom of complex numbers while using three color channels (L, A, B) that can be processed by standard color space conversion algorithms, thus bridging the gap between mathematical representation and processing requirements.
2Adaptability or versatility
If the number of degrees of freedom per pixel is increased to match RGB color space (three degrees), then color space conversion becomes more compatible, but information loss occurs when converting from complex numbers (two degrees of freedom) to three-color values
Solution Approach 1:
The patent changes the parameter mapping approach by using the real part for L channel and imaginary part for A and B channels, creating a lossless transformation that maintains the original information while achieving compatibility with three-channel color space processing.
Solution Approach 2:
The patent effectively adds a dimensional transformation by mapping two-dimensional complex number space into three-dimensional LAB color space through a specific projection method that preserves all original information. This dimensional change enables compatibility with standard color processing while maintaining information integrity.
Data Source
AI summary
A signal processing device includes: a window function unit that performs window function processing on a Fourier transform frame cut from time-series signal data; a Fourier transform unit that transforms, into information in a frequency domain, the Fourier transform frame subjected to the window function processing; a color space conversion unit that generates spectrogram data by converting a phase angle of a complex amplitude in the frequency domain into a hue and performing conversion that associates an absolute value of the complex amplitude with brightness by mapping, the spectrogram data being defined so as to correspond to a two-dimensional plane of time and frequency; and a deep learning processing unit that detects or identifies a signal by convolutional processing and fully connected processing on the spectrogram data.