Anomaly detection device and method using neural network, and device and method for training neural network
An anomaly detection device according to one embodiment may comprise: a receiver for receiving a hyperspectral image; and a processor for extracting, on the basis of a plurality of target partial autoencoders corresponding to a plurality of bands included in the hyperspectral image, a plurality of local features corresponding to the plurality of bands, extracting a global feature on the basis of the plurality of local features through an aggregate autoencoder, and detecting anomalies on the basis of the global feature.
1 . An anomaly detection device comprising:
a receiver configured to receive a hyperspectral image; and
a processor configured to detect outliers in the hyperspectral image based on a pretrained target neural network,
wherein the pretrained target neural network comprising,
a target partial encoder configured to which each window generated from the hyperspectral image is input;
an aggregate autoencoder configured to which each output of the target partial encoder is connected and input;
and a target partial decoder configured to which output of the aggregate autoencoder is divided and input,
wherein the processor is configured to divide the hyperspectral image into the window corresponding to a plurality of bands using a sliding window,
wherein the processor is configured to:
extract a first local feature by inputting a first window corresponding to a first band among the plurality of bands to a first partial autoencoder;
extract a second local feature by inputting a second window corresponding to a second band among the plurality of bands to a second partial autoencoder;
generate an aggregate latent vector by connecting a plurality of local features;
extract a global feature by inputting the aggregate latent vector to an encoder of the aggregate autoencoder;
generate aggregate output by inputting the global feature to a decoder of the aggregate autoencoder;
generate a plurality of divided aggregate outputs by dividing the aggregate output; and
generate a plurality of restored images by inputting the plurality of divided aggregate outputs to the target partial decoder.