Application of neural network technology to fiber image analysis, final report: analytical methods for asbestos fibers.
Deye GJ; Baron PA
NIOSH 1990 Dec; :1-43
In an attempt to provide improved detection of asbestos (1332214) fibers over traditional image analysis techniques, neural network technology was applied to the phase contrast light microscopy detection of fiber images. The available computer power limits the information that can be extracted by the neural network. Subimages containing detected fiber segments were used to train the network. The network was expected to ensure that each segment was part of a fiber and to extend the segment if it was not completely detected. The goal of the operation was to improve the detection of fibers so that more accurate fiber counts could be obtained and to improve the measurement of fiber length. While the network did make some improvements in the image, the error rate was unacceptably high. A high false positive rate indicated that too many nonfibrous features were detected as fibers. A high false negative rate indicated that too many real fibers were not detected. Suggestions are made to improve the system further.
NIOSH-Author; Analytical-chemistry; Analytical-methods; Chemical-analysis; Asbestos-fibers; Biological-monitoring; Fibrous-dusts
NTIS Accession No.
Monitoring and Control Research Branch, Division of Physical Sciences and Engineering, NIOSH, Cincinnati, Ohio, 43 pages, 14 references