Electromagnetic Reflection Imaging With Segmented Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection techniques using electromagnetic waves face challenges in reducing data amount while maintaining estimation accuracy, leading to reduced image sharpness and difficulty in manual labeling operations.
Innovation Solution
A processing apparatus and method that generates label determination and learning images using partial signals of reflection waves, associates these with labels to create training data, and employs machine learning to develop an estimation model for accurate object detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If the amount of electromagnetic wave signal data is reduced to lower processing load and reduce the number of transmission/reception antennas, then processing load is reduced and device size is miniaturized, but image sharpness deteriorates to a level where humans cannot recognize objects
Solution Approach 1:
The patent segments the signal data into two distinct types: training data signals (first signals) used for generating estimation models, and estimation data signals (second signals) used for actual object detection. This segmentation allows the system to use high-quality training data with sufficient sharpness for model generation while using reduced-data estimation data for actual detection operations, thereby resolving the contradiction between device miniaturization and image sharpness requirements.
Solution Approach 2:
The patent performs preliminary training model generation using high-quality training data before actual object detection operations. The estimation model is pre-trained on comprehensive training data that maintains high image sharpness, allowing the actual detection system to operate with reduced sensor complexity while still achieving accurate object recognition through the pre-learned model.
2Productivity
If the amount of electromagnetic wave signal data is reduced to achieve miniaturization and lower processing load, then processing load is reduced, but manual labeling operation becomes difficult due to insufficient image quality
Solution Approach 1:
The patent segments data usage into training phase (using comprehensive high-quality data) and estimation phase (using reduced data). During training, sufficient data maintains image sharpness enabling manual labeling. During estimation, reduced data lowers processing load. This temporal segmentation resolves the contradiction between productivity and ease of operation.
Solution Approach 2:
The system performs preliminary training with comprehensive data to create robust estimation models before actual operation. This preliminary action establishes the computational intelligence needed to perform accurate object detection and labeling automatically, reducing or eliminating the need for manual labeling operations while maintaining processing efficiency.
3Loss of time
If the number of transmission antennas is reduced to shorten irradiation time and suppress motion blur, then irradiation time is shortened and motion blur is suppressed, but image sharpness deteriorates
Solution Approach 1:
The patent applies segmentation by distinguishing between training data acquisition (which can use longer irradiation times with more antennas for high sharpness) and actual estimation operations (which use fewer antennas and shorter times). The training phase builds comprehensive models with high sharpness images, while the estimation phase benefits from faster, blur-reduced capture without sacrificing detection accuracy.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Reduces processing load, miniaturizes sensors, and maintains estimation accuracy by generating images with sufficient sharpness for both human and computer recognition, facilitating efficient labeling operations.
Implementation Method 1
an electromagnetic wave transmission/reception means for irradiating an electromagnetic wave from a transmission antenna, and receiving a reflection wave by a reception antenna
Data Source
AI summary
The present invention provides a processing apparatus (10) including an electromagnetic wave transmission/reception unit (11) that irradiates an electromagnetic wave from a transmission antenna, and receives a reflection wave by a reception antenna; a label determination image generation unit (12) that generates a label determination image, based on a signal of the received reflection wave; a learning image generation unit (13) that generates a learning image, based on a signal being a part of a signal of the receive reflection wave, and less than a signal to be used in generation of the label determination image; a label determination unit (14) that determines a label, based on the label determination image; and a training data generation unit (15) that generates training data in which the learning image and the label are associated, and causing a training data storage unit (16) to store the generated training data.