Learning-Based See-Through Sensing for Layered Image Reconstruction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current see-through sensing technologies, such as terahertz sensing, face challenges in reconstructing three-dimensional object images due to complexity and quality degradation caused by the shadow effect from non-uniform wave penetration through layered structures, leading to artifacts in deeper layers.

Innovation Solution

A scanner system that treats the target object as a multi-layer structure, using a neural network to classify waves modified by penetration, emitting waves in parallel directions to penetrate layers, and measuring intensities to reconstruct images of each layer with pixel values based on class labels, addressing the shadow effect by considering unique modifications for classification-based image reconstruction.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional image reconstruction techniques are used for three-dimensional objects, then the structure can be visualized, but computational complexity increases and image quality degrades due to shadow effects

Engineering Contradiction:
Improveimage reconstruction qualityVSAvoidcomputation complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent divides the three-dimensional object into multiple two-dimensional layers along the wave propagation direction. Each layer is reconstructed independently using 2D sparse reconstruction algorithms, avoiding the computational burden of 3D reconstruction while maintaining image quality. The segmentation approach processes each layer separately, reducing the overall computational complexity.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent transforms the 3D image reconstruction problem into a series of 2D reconstruction problems by slicing the object along the propagation direction. This dimensionality reduction from 3D to 2D simplifies the computational task while preserving the essential structural information through the layered representation.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Difficulty of detecting and measuring

If waves penetrate through layered structures, then internal structure can be detected, but non-uniform penetration causes shadow effects and artifacts in deeper layers

Engineering Contradiction:
Improveinternal structure detectionVSAvoidimage accuracy
Core Design Contradiction:
Difficulty of detecting and measuringVSReliability

Solution Approach 1:

The patent applies different processing approaches to different regions of the layered structure. For each layer, the reconstruction algorithm accounts for the local attenuation characteristics and shadow effects specific to that depth level. This localized processing improves accuracy by adapting to the varying penetration conditions at different depths.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent converts the shadow effect, traditionally considered a harmful artifact, into useful information. By modeling the shadow effects caused by upper layers, the algorithm can compensate for attenuation and enhance the visibility of deeper structures. The shadow patterns provide information about the intervening layers that can be used to improve overall reconstruction accuracy.

Inventive Principle:
Principle #22Blessing in disguise (Convert harm into benefit)

3Measurement precision

If hierarchical image recovery is used to increase image quality, then artifacts are reduced, but the reconstruction process becomes more complex

Engineering Contradiction:
Improvereconstructed image qualityVSAvoidreconstruction technique complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent implements hierarchical image recovery by segmenting the reconstruction process into multiple levels corresponding to different depth layers. Each layer is reconstructed with appropriate detail level, and the results are combined to form the complete 3D structure. This segmented approach achieves high image quality while keeping each individual reconstruction step manageable in complexity.

Inventive Principle:
Principle #1Segmentation

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

This approach improves the computational efficiency and accuracy of image reconstruction by leveraging the unique modification of waves through layered structures, reducing artifacts and enhancing image quality, especially in factory automation and real-time diagnostics.

Implementation Method 1

an emitter configured to emit a set of waves in parallel directions of propagation to penetrate a sequence of layers of the target object

Methodology Applied
Scientific EffectWave propagation:

Implementation Method 2

a receiver configured to measure intensities of the set of waves modified by penetration through the layers of the target object

Methodology Applied
Scientific EffectWave intensity measurement:

Data Source

PatentEP4022341B1Learning-based see-through sensing suitable for factory automation
Publication Date: 2024.08.07 MITSUBISHI ELECTRIC CORP
  • EP4022341B1 patent drawingFigure 1
  • EP4022341B1 patent drawingFigure 2
  • EP4022341B1 patent drawingFigure 3

AI summary

A scanner for image reconstruction of a structure of a target object uses a neural network trained to classify each segment of a sequence of segments of a modified wave into one or multiple classes. The sequence of segments corresponds to the sequence of layers of the target object, such that a segment of modified wave corresponds to a layer having the same index in the sequence of layers as an index of the segment in the sequence of segments. The scanner executes the neural network for each wave modified by penetration through the layers of the target object to produce the classes of segments of the modified waves. Next, the scanner selects the classes of segments of different modified waves corresponding to the same layer to produce an image of the layer of the target object with pixel values being functions of labels of the selected classes.