Helmut Strecker

Automated X-ray inspection of aluminum castings

The experience gained with several approaches to automatic flaw detection in X-ray images of cast aluminum wheels is described. Basic problems are mentioned, and the applicability of segmentation methods to actual inspection tasks is demonstrated. The discussion focuses on the definition, extraction, and combination of local features for pixel classification. Results of pilot tests are described briefly. Further investigations are suggested, aiming at more generality of the methods and greater stability of the segmentation. >

Detection of explosives in airport baggage using coherent x-ray scatter

Bulk objects can be investigated for their material constituents by applying high-energy (30 keV to 100 keV) coherent X-ray scattering. When aiming at the detection of explosives in airport baggage, the technique allows discrimination between explosives and other substances. Coherent X-ray scatter measurements are presented for a set of explosives and their constituents as well as for a variety of nonexplosive materials. They demonstrate the superior material discrimination power of this method. The measurements have provided a quantitative basis for the prototype design of an airport baggage scanner. Sensitivity (200 g) and inspection time requirements (a few seconds) demand a highly application-specific system design with parallel acquisition and analysis of scatter spectra from different volume elements.

Configuration Using PLAKON - An Applications Perspective

PLAKON is a tool for building Planning and Configuration Expert Systems in technical domains. It is currently being developed and tested within the joint BMFT-project TEX-K. In this paper, an overview of the PLAKON software components is given. The basic mechanisms are explained and exemplified by a typical configuration application, the configuration of Automatic X-ray Inspection Systems. Emphasis is laid on the definition of a taxonomically- and compositionally-structured conceptual hierarchy with formulation of constraints and control knowledge.

XRAY - An Experimental Configuration Expert System for Automatic X-ray Inspection

An experimental expert system for knowledge based configuration in automatic X-ray inspection is presented. The system is able to select, arrange and adapt image analysis operators according to a given inspection task. Configured analysis sequences are presented, which automatically detect and classify flaws in cast aluminium parts. Planned developments are briefly mentioned.

Simulation-based training and testing of classification schemes for CXRS explosives detection

In explosives detection with coherent x-ray scattering, radiation scattered in the forward direction at a small fixed angle is accumulated from volume voxels of interest with an energy- resolving multichannel detector. For most military and industrial explosives, the resulting scatter spectra exhibit characteristic diffraction peaks which correspond to their polycrystalline structure. The performance of a spectrum processing and classification scheme can be analyzed by simulation. The simulation is based on spectra measured with good counting statistics. It allows the influence of the following parameters or effects to be studied: 1) total photon flux, responsible for the quantum noise; 2) arbitrary mixing of the measured substances and partial volume effects; and 3) attenuation of the primary and scatter radiation.

Lernverfahren zur Segmentierung industrieller Röntgenbilder

Die Segmentierung von Rontgenbildern industrieller Objekte wird durch das Lernen von Pixel-Klassifikatoren auf der Basis lokaler Merkmale versucht. Dazu werden Lernstichproben durch interaktives Markieren reprasentativer Bildpunkte mit ihrem Klassenindex erstellt. Ein Polynomansatz verknupft ausgewahlte, vorzugsweise rotationsinvariante lokale Merkmale zu Trennfunktionen, deren Parameter aus einer Quadratmittelanpassung an die Lernstichprobe gewonnen werden. Erste Versuche zur Erkennung von Lunkern in Guβteilen liefern Trennfunktionen, in denen die Lunker hell hervorgehoben werden, fuhren aber noch nicht zu befriedigenden Segmentierungsresultaten. Verbesserungen werden durch Verwendung seguentieller Pixelklassifikation mit auf Zernike-Polynomen basierenden Merkmalssatzen angestrebt.