Browsing by Author "Chacon, M"

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    Automated testing of aluminum castings using classifier fusion strategies
    (AMER SOC NONDESTRUCTIVE TEST, 2005) Mery, D; Chacon, M; Gonzalez, L; Munoz, L
    Generally, discontinuity detection in automated visual testing consists of two steps: identification of potential discontinuities using image processing techniques and classification of potential discontinuities into discontinuities and regular structures (false alarms) using a pattern recognition methodology. In the second step, since several features cyan be extracted from the potential discontinuities, a feature selection must be performed. In this paper, several known classifiers are studied in automated visual testing: threshold, euclidean, mahalanobis, polynomial, support vector machine and neural network classifiers. First, the performance of the classifiers is assessed individually. Second, the classifiers are combined in order to improve their performance. Seven fusion strategies in the combination were tested: and, or, majority vote, product, sum, maximum and median. The proposed methodology was tested on real data acquired from 50 noisy radiographic images of aluminum wheels, where 23 000 potential discontinuities (with only 60 real discontinuities) were segmented and 405 features were extracted for each potential discontinuity. Using fusion of classifiers, a very good performance was achieved, yielding a sensitivity of 100% and specificity of 99.97%.
  • Thumbnail Image
    Item
    Neural network method for failure detection with skewed class distribution
    (BRITISH INST NON-DESTRUCTIVE TESTING, 2004) Carvajal, K; Chacon, M; Mery, D; Acuna, G
    The automatic detection of flaws through non-destructive testing uses pattern recognition methodology with binary classification. In this problem a decision is made about whether or not an initially segmented hypothetical flaw in an image is in fact of law. Neural classifiers are one among a number of different classifiers used in the recognition of patterns. Unfortunately, in real automatic flaw detection problems there are a reduced number of flaws in comparison with the large number of non-flatus. This seriously limits the application of classification techniques such is artificial neural networks elite to the imbalance between classes. This work presents a new methodology for efficient training with imbalances in classes. The premise of the present work is that if there are sufficient cases of the smaller class, then it is possible to reduce the Size of the larger class by using the correlation between cases of this latter class, with a minimum information loss. It is then possible to create it training set for a neural model that allows good classification. To test this hypothesis a problem of great interest to the automotive industry is used, which is the radioscopic inspection of cast aluminium pieces. The experiments resulted in perfect classification of 22936 hypothetical flaws, of which only 60 were real flat-vs and the rest were false alarms.