.png){kind=icon}
.png){kind=icon}
Return
Segment-driven anomaly detection in hyperspectral data using watershed technique
.png){kind=icon}
OA
AI
DOI：10.1016/j.ejrs.2024.03.007.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
A significant portion of hyperspectral image (HSI) analysis involves detecting anomalous pixels, which are indicative of interesting phenomena or objects. One of the main challenges is the presence of outlier and noisy pixels in background data due to the variety of spectral signatures in heterogeneous HSIs. This article presents an effective approach using both spectral and spatial features for anomaly detection. The median filter with an appropriate size driven by using the principal component information is used for cleaning the background. Then, the image is segmented using the watershed approach. The anomaly detection occurs based on the spatial resolution by calculating each pixel ' s distance from its segment via spectral angle or Euclidean distance. The proposed Watershed Anomaly Detector (WAD), employs spatial features to segment the HSI properly. It also uses spectral features within each segment to detect anomalous pixels. The WAD outperforms other methods due to its simplicity and conceptual clarity. Notably, its underlying equation offers broader applicability for HSI segmentation tasks. Experiments on three benchmark datasets show WAD achieves higher accuracy and faster execution versus state-of-the-art techniques. On average across the datasets and methods, WAD attained a 6.45% higher area under the receiver operating characteristic (ROC) curve and ran 26.95 s faster than other detectors. The WAD effectively detects anomalies in varied spectral and spatial resolutions. The results highlight the stability, robustness and computational efficiency of the proposed approach across diverse data. The simultaneous effectiveness and efficiency make WAD well-suited for near real -time anomaly detection applications.
Keywords:
Hyperspectral image
Anomaly detection
Segmentation
Spectral and spatial features
Watershed
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
E
IF:
4.1
Papers:
491
Citations:
2.9K