Gaussian Mixture Feature Improved Autoencoder for FOD Detection on Airport Runway

Image reconstruction based on autoencoder can detect unknown types of foreign object debris on airport runway. To tackle the problem of anomaly reconstruction caused by the strong generalization capability of autoencoder, the Gaussian mixture feature improved autoencoder is proposed. This model first obtains the encoding features of the input image through the encoder, and then the Gaussian mixture model identifies and eliminates potential abnormal features for reconstruction. At the training stage, the estimation network is combined to perform end-to-end optimization of autoencoder and Gaussian mixture model to build an accurate normal model. At the test stage, the reconstructed image is obtained from the encoding feature that removes abnormal information, which will tend to be close to the normal image, resulting in a large reconstruction error for anomalies. Furthermore, the multiple hierarchical semantic difference is employed to improve the accuracy of the anomaly detection. The experimental validation is carried out on the runway pavement dataset FOD AD from an airport in China and the public anomaly detection dataset MVTec AD. The experimental results demonstrate that the average AUC of the proposed model on FOD AD and MVTec AD achieves 0.981 and 0.985, respectively, which are better than similar methods in comparison.