Abnormal sound detecting device, abnormality model learning device, abnormality detecting device, abnormal sound detecting method, abnormal sound generating device, abnormal data generating device, abnormal sound generating method, and program
A technology of abnormal sound and detection device, which is applied in the field of abnormal detection, can solve the problems of cost and other problems, and achieve the effect of improving accuracy
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
no. 1 approach ＞
[0036] In the present embodiment, an abnormality model (or penalty) is estimated from a small amount of abnormal sound data, and the abnormality degree is calculated while using it in combination. Kernel density estimation is used in the estimation of the anomaly model. In the present invention, instead of estimating an abnormality model using mixed weights that are equal to all abnormal sound data as in the conventional kernel density estimation method, the abnormality model is calculated using weights that maximize the accuracy of abnormal sound detection. In order to realize it, in the present invention, realize the algorithm that weight is optimized, so that under the abnormal judgment threshold value that can judge all the abnormal sound data that obtains as abnormal, make normal sound data misjudged as abnormal probability (false positive) class rate) minimum.
[0037]
[0038] refer to figure 1, illustrating previous unsupervised anomalous sound detection techniques...
Embodiment approach
[0094] Hereinafter, the first embodiment of the present invention will be described in detail. The first embodiment of the present invention is composed of an abnormality model learning device 1 for estimating an abnormality model used in abnormal sound detection, and an abnormal sound determination device 2 for judging whether an observed signal is normal or abnormal using the abnormality model learned by the abnormality model learning device 1 .
[0095] "Abnormal Model Learning Device"
[0096] Such as Figure 5 As exemplified in , the abnormality model learning device 1 of the first embodiment includes: an input unit 11 , an initialization unit 12 , a threshold determination unit 13 , a weight update unit 14 , a weight correction unit 15 , a convergence determination unit 16 , and an output unit 17 . The abnormal model learning device 1 performs Figure 6 The processing of each step exemplified in the above implements the abnormal model learning method of the first embo...
no. 2 approach ＞
[0121] In this embodiment, a framework for improving the accuracy of unsupervised abnormal sound detection is provided by using a small amount of abnormal sound data obtained. In the present embodiment, an abnormality model (or penalty) is estimated from a small amount of abnormal sound data, and the abnormality degree is calculated while using them together. The anomaly model is defined as the similarity between a small number of anomalous sounds and observed signals. That is, a penalty is given that is easily determined to be abnormal for an observation signal similar to the abnormal sound obtained so far. In order to learn the abnormal model, an algorithm to optimize the weight is provided, so that under the abnormal judgment threshold that can judge all abnormal data as abnormal, the probability of normal observation signals being misjudged as abnormal, that is, the false positive rate is the smallest.
[0122]
[0123] Abnormal sound detection is a task of judging whet...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More