AI Classifier Training Using Synthetic EM Scan Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current AI classifiers used in electromagnetic screening for detecting concealed objects-of-interest (OOI) can introduce unintended bias due to human selection of training images, leading to suboptimal performance in object detection and classification.
Innovation Solution
The method involves training an AI classifier using a diverse set of raw electromagnetic (EM) scans, including both in-class and not-in-class samples, with iterative adjustments to parameters based on error measurement, and employing a combination of raw EM sample-trained and conventional visible pixel array-trained classifiers for improved object classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If human-selected training images are used to train AI classifiers, then the training process can be completed with available data, but the classifier introduces unintended bias leading to suboptimal detection performance
Solution Approach 1:
The patent uses synthetic EM scan data generated from 3D models and physical property databases to create training samples that replicate real-world scenarios without relying on human-selected images. This copying approach from simulated to real domains provides diverse, bias-free training data while maintaining physical accuracy through electromagnetic property modeling
Solution Approach 2:
The system varies multiple parameters in synthetic data generation including electromagnetic properties, object orientations, positions, and scan configurations to create comprehensive training sets. This parameter variation ensures the classifier learns robust features across diverse conditions without human bias in sample selection
2Ease of manufacture
If manual selection of training samples is performed, then the process can be completed with human judgment, but it requires significant effort and time
Solution Approach 1:
The system automatically generates synthetic training data through computer-generated 3D models, electromagnetic property databases, and automated scan simulation. This self-service approach eliminates manual sample selection entirely, requiring no human time or effort while producing comprehensive, diverse training datasets
Solution Approach 2:
The patent pre-computes electromagnetic properties, creates 3D object models, and generates synthetic scan data before the actual classification task. This preliminary action of preparing synthetic training data in advance eliminates the need for time-consuming manual sample selection during deployment
3Productivity
If conventional classifiers with manual feature extraction are used, then the system can operate with traditional methods, but algorithm optimization requires significant manual effort
Solution Approach 1:
The patent replaces manual feature extraction and classifier configuration with end-to-end deep learning models that automatically learn optimal features from raw EM scan data. This substitution eliminates manual algorithm configuration while achieving superior classification performance through automated neural network training on synthetic data
Solution Approach 2:
The deep learning model serves multiple functions simultaneously: it performs feature extraction, classification, and optimization in a single unified architecture. This multi-functionality replaces the need for separate manual configuration steps for each classification task
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 reduces unintended bias and enhances the accuracy of AI classifiers in distinguishing between OOI and non-OOI, leading to improved detection and classification performance in electromagnetic screening processes.
Implementation Method 1
EM scanning a subject, including transmitting an EM scan energy toward the subject, receiving an EM return
Data Source
AI summary
Systems and methods are described, and an example method includes a training an artificial intelligence (AI) classifier of scanned items, including obtaining a training set of sample raw scans. The set includes a population of sample in-class raw scans, which include blocks of sensor data from scans of regions having in-class objects, and the set includes a population of sample not-in-class raw scans, which include blocks of sensor data from scan of regions without in-class objects. The example includes applying the AI classifier to sample raw scans in the training set, measuring errors in the results, and updating classifier parameters based on the errors, until detecting a training completion state.