Multi-Energy CT Object Classification via Local Density Distribution Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current baggage scanning systems, particularly those using X-ray computed tomography (CT) technology, face challenges in detecting thin sheet explosives due to their orientation and low density contrast with the background, leading to high false alarm rates and missed detections.
Innovation Solution
The method employs multi-energy CT scanners to analyze local distribution features of density and atomic number data, using first and second-order statistics from CT and Z images to classify objects, dividing objects into core and surface portions, and computing histograms to enhance discrimination capabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional X-ray CT scanning is used, then scanning speed and coverage are maintained, but detection accuracy for thin sheet explosives deteriorates due to low density contrast and orientation dependence
Solution Approach 1:
The patent segments the CT image data into multiple energy levels (first energy and second energy images) and further divides the object analysis into core portions and surface portions. This segmentation allows independent analysis of different regions and energy levels, improving detection accuracy for thin sheets while maintaining scanning efficiency through parallel processing of segmented data.
Solution Approach 2:
The patent introduces a new dimension of analysis by computing local distributions of density and atomic number across multiple energy levels. Instead of relying solely on single-energy density contrast, the system analyzes the distribution characteristics in the energy dimension, enabling detection of thin sheets with low density contrast through their distinctive local distribution patterns.
2Measurement precision
If single-energy CT imaging is used, then image acquisition is simple and fast, but discrimination capability between different materials deteriorates
Solution Approach 1:
The patent segments the imaging data into multiple energy levels and further divides object analysis into core and surface portions. This segmentation enables independent processing of each segment, reducing overall processing complexity while enhancing material discrimination through multi-energy analysis of distributed regions.
Solution Approach 2:
The patent applies local quality analysis by computing local distributions of density and atomic number for different portions (core and surface) of objects. This allows the system to capture material-specific characteristics in different regions, improving discrimination capability without requiring complex global analysis of the entire image.
3Reliability
If threshold-based object classification is used, then processing speed is maintained, but false alarm rate increases due to insufficient material characterization
Solution Approach 1:
The patent performs preliminary analysis by computing local distributions and statistical characteristics (mean, standard deviation) of density and atomic number before final classification. This preliminary characterization of material properties enables more accurate threshold-based classification, reducing false alarms while maintaining processing speed through efficient statistical computation.
Solution Approach 2:
The patent introduces local distribution statistics as an intermediary between raw CT data and final classification decisions. By computing statistical characteristics of local regions, the system creates a richer feature representation that improves classification accuracy without requiring complex real-time analysis, thus maintaining processing speed while reducing false alarms.
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 false alarm rates by providing richer information about objects, enabling more accurate detection and classification of threats, including thin sheet explosives, by analyzing local density and atomic number distributions within CT images.
Implementation Method 1
measurements of the local distributions of density and atomic number of portions of objects in the objects from CT images and Z images generated by a multi-energy computed tomography scanner
Data Source
AI summary
A method of and a system for identifying objects using local distribution features from multi-energy CT images are provided. The multi-energy CT images include a CT image, which approximates density measurements of scanned objects, and a Z image, which approximates effective atomic number measurements of scanned objects. The local distribution features are first and second order statistics of the local distributions of the density and atomic number measurements of different portions of a segmented object. The local distributions are the magnitude images of the first order derivative of the CT image and the Z image. Each segmented object is also divided into different portions to provide geometrical information for discrimination. The method comprises preprocessing the CT and Z images, segmenting images into objects, computing local distributions of the CT and Z images, computing local distribution histograms, computing local distribution features from the said local distribution histograms, classifying objects based on the local distribution features.