Method for detecting and classifying anomalies using artificial neural networks

• US 6,622,135 B1
• Filed: 12/16/1999
• Issued: 09/16/2003
• Est. Priority Date: 12/29/1998
• Status: Expired due to Term

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: International Business Machines Corporation
• Inventors: Williams, Erin, Imbert De Tremiolles, Gislain, Tannhof, Pascal
• Primary Examiner(s): STARKS, WILBERT L
• Application Number:
  US09/465,088
• Time in Patent Office: 1,370 Days
• Field of Search: 706/20, 706/25, 382/128, 382/149, 382/159, 382/225, 340/572
• US Class Current: 706/20
• CPC Class Codes: G06F 18/24147 Distances to closest patterns, e.g. nearest neighbour classification
• CPC Class Codes: G06T 2207/30148 Semiconductor; IC; Wafer
• CPC Class Codes: G06T 7/0004 Industrial image inspection

Abstract

To avoid the problem of category assignment in artificial neural networks (ANNs) based upon a mapping of the input space (like ROI and KNN algorithms), the present method uses “probabilities”. Now patterns memorized as prototypes do not represent categories any longer but the “probabilities” to belong to categories. Thus, after having memorized the most representative patterns in a first step of the learning phase, the second step consists of an evaluation of these probabilities. To that end, several counters are associated with each prototype and are used to evaluate the response frequency and accuracy for each neuron of the ANN. These counters are dynamically incremented during this second step using distances evaluation (between the input vectors and the prototypes) and error criteria (for example the differences between the desired responses and the response given by the ANN). At the end of the learning phase, a function of the contents of these counters allows an evaluation of these probabilities for each neuron to belong to predetermined categories. During the recognition phase, the probabilities associated with the neurons selected by the algorithm permit the characterization of new input vectors and more generally any kind of input (images, signals, sets of data) to detect and classify anomalies. The method allows a significant reduction in the number of neurons that are required in the ANN while improving its overall response accuracy.

Claims

1. A method of detecting anomalies and classifying the anomalies in categories using an artificial neural network (ANN) comprising the steps of:

performing a preliminary learning phase comprising;

creating a blank block, the blank block having a shape that is designed to view each set of data in its totality in a determined number of steps;

stepping each blank block through each image;

generating an input block derived from said blank block for each step where each input block is a binary photograph of the image at each step, each input block having a central point;

storing the representative input blocks for at least a number of steps,

creating a first database using the stored representative input blocks, the first database defining prototypes of a first ANN,

providing a recognition phase comprising;

calculating the probability of each of said prototypes belonging to defined categories;

repeating at least once said blank block stepping through each set of data;

using said prototypes and their associated probabilities to characterize new subsets of data, wherein each subset of data is characterized by subset probabilities, the subset probabilities selected from the group consisting of the K nearest neighbor (KNN) algorithm and a function of KNN;

replacing the central point of the input blocks with the subset probabilities.

2. An artificial neural network (ANN) comprised of a plurality of neurons implementing a mapping of the input space comprising:

a storage means for storing prototype components in each neuron;

a plurality of counting means to compute the distances between an input vector presented to the ANN and the prototypes stored in each neuron; and

,

a specific storage means for storing data relative to the neuron behavior.

3. The ANN of claim 2 wherein at least two of said counting means comprise two counters, a first counter to count the neuron'"'"'s response accuracy and a second counter to count the neuron'"'"'s response frequency.