| CPC G06T 7/0002 (2013.01) [G06F 18/22 (2023.01)] | 8 Claims |
|
1. An anomaly detection device comprising:
at least one hardware processor configured to execute processes comprising:
calculating a difference in feature map output in an intermediate layer when each of one or more pieces of normal data included in a normal dataset and input data are input to a pre-trained convolution neural network (CNN) as an anomaly score map indicating a degree of anomaly of the input data, the input data and the normal data being photographed image data;
selecting, as reference data, one piece of the normal data which is included in the normal dataset and which is determined to have a photographing condition whose degree of similarity with a photographing condition of the input data is equal to or greater than a predetermined threshold value;
by using a difference in feature within the one piece of the normal data selected as the reference data, calculating a correction score map for correcting the anomaly score map, the calculated correction score map indicating a higher score as the difference in feature is greater;
correcting the anomaly score map by using the correction score map; and
displaying, on a display, a corrected anomaly score map that represents the anomaly score map corrected by using the correction score map.
|