An Innovative Application of Isolation-Based Nearest Neighbor Ensembles on Hyperspectral Anomaly Detection
- 11 June 2024
- journal article
- research article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Geoscience and Remote Sensing Letters
- Vol. 21 (1545598X) , 1-5
Abstract
This letter presents an innovative application of the isolation using nearest neighbor ensemble (iNNE) method for hyperspectral anomaly detection. iNNE is an efficient anomaly detector based on nearest neighbors and isolation. Based on iNNE, we propose a novel isolation-based hyperspectral anomaly detection framework that detects anomalies as follows. First, the Gabor filter is applied to extract spatial information from the PCA-projected subspace. Gabor features are then employed as the input to the ReMass-iForest to obtain the spatial anomaly score. Then the iNNE is employed to derive the spectral anomaly score. Finally, we integrate the detection outcomes by linearly combining the acquired spatial and spectral anomaly scores to predict anomaly pixels. Compared with the existing tree-based isolation method, such as the iForest, the proposed iNNE-based method addresses the poor performance of iForest in detecting local anomalies and anomaly detection in high-dimensional data. Experimental analysis across five hyperspectral datasets indicates that the proposed detector consistently attains the highest AUC scores and exhibits improved ROC curves in comparison to established benchmarks.Keywords
Funding Information
- Science and Technology Research and Development program of china state railway group (K2023G032)
- Science and Technology Research and Development program of china state railway group (Q2023G022)
- Science and Technology Research and Development Program of China Railway Design Corporation (2023A0253820)
- Science and Technology Research and Development Program of China Railway Design Corporation (2021A241010)
This publication has 21 references indexed in Scilit:
- Hyperspectral Anomaly Detection Using Attribute ProfilesIEEE Geoscience and Remote Sensing Letters, 2017
- Hyperspectral Anomaly Detection by the Use of Background Joint Sparse RepresentationIEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2015
- Efficient Anomaly Detection by Isolation Using Nearest Neighbour Ensemble2014
- Weighted-RXD and Linear Filter-Based RXD: Improving Background Statistics Estimation for Anomaly Detection in Hyperspectral ImageryIEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2014
- On Detecting Clustered Anomalies Using SCiForest2010
- Isolation Forest2008
- A support vector method for anomaly detection in hyperspectral imageryIEEE Transactions on Geoscience and Remote Sensing, 2006
- A cluster-based approach for detecting man-made objects and changes in imageryIEEE Transactions on Geoscience and Remote Sensing, 2005
- Kernel RX-algorithm: a nonlinear anomaly detector for hyperspectral imageryIEEE Transactions on Geoscience and Remote Sensing, 2005
- Adaptive multiple-band CFAR detection of an optical pattern with unknown spectral distributionIEEE Transactions on Acoustics, Speech, and Signal Processing, 1990