Exploring Hyperspectral Anomaly Detection with Human Vision: A Small Target Aware Detector

TL;DR

This work targets hyperspectral anomaly detection (HAD) by moving detection into a perceptual feature space rather than the reconstruction/error space. It introduces STAD, a small-target aware detector that uses saliency maps to reveal anomalous representations, a small-target filter to suppress low-confidence regions, and a teacher–student distillation framework to enable lightweight deployment on edge devices. Empirical results on the HAD100 dataset show STAD achieving top-level performance in both anomaly detection accuracy (AUC) and background suppression, with strong robustness across 94 HSIs. The study demonstrates the value of feature-space representations aligned with human visual perception for HAD and provides a practical, interpretable approach for real-time hyperspectral anomaly localization.

Abstract

Hyperspectral anomaly detection (HAD) aims to localize pixel points whose spectral features differ from the background. HAD is essential in scenarios of unknown or camouflaged target features, such as water quality monitoring, crop growth monitoring and camouflaged target detection, where prior information of targets is difficult to obtain. Existing HAD methods aim to objectively detect and distinguish background and anomalous spectra, which can be achieved almost effortlessly by human perception. However, the underlying processes of human visual perception are thought to be quite complex. In this paper, we analyze hyperspectral image (HSI) features under human visual perception, and transfer the solution process of HAD to the more robust feature space for the first time. Specifically, we propose a small target aware detector (STAD), which introduces saliency maps to capture HSI features closer to human visual perception. STAD not only extracts more anomalous representations, but also reduces the impact of low-confidence regions through a proposed small target filter (STF). Furthermore, considering the possibility of HAD algorithms being applied to edge devices, we propose a full connected network to convolutional network knowledge distillation strategy. It can learn the spectral and spatial features of the HSI while lightening the network. We train the network on the HAD100 training set and validate the proposed method on the HAD100 test set. Our method provides a new solution space for HAD that is closer to human visual perception with high confidence. Sufficient experiments on real HSI with multiple method comparisons demonstrate the excellent performance and unique potential of the proposed method. The code is available at https://github.com/majitao-xd/STAD-HAD.

Figures (8)

• Figure 1: One of the results of testing on the HAD100 dataset. (a) Result of reconstruction-based method, the anomalies within the red circles are not detected because of their easy reconstruction. (b) Results of the saliency map based method, more anomalies are detected compared to reconstruction-based method. (c) Anomalies map.
• Figure 2: Schematic representation of the anomalous representation ability of feature maps and reconstruction error maps. A set of maps with the same color border represents the label (bottom row), feature map (solid box), and reconstruction error map (dashed box) of the same HSI. Note that the feature maps are obtained by solving for the maximum value of the channel dimension, which does not imply that the detection results are so. The reconstruction errors are calculated by mean square error (MSE).
• Figure 3: Schematic of our proposed approach. (a) The training process of the teacher network. (b) The process of training student network through knowledge distillation. (c) Obtaining the mask matrix through STF. (d) Detecting and localizing anomalous targets in HSI by STAD.
• Figure 4: Feature maps of the intermediate layers of the fully connected layer network (a) and the convolutional layer network (b), and (c) is the anomaly map.
• Figure 5: One of the results of testing on the HAD100 dataset. (a) Result of estimation-based method. (b) Results of the saliency map based method, more anomalies are detected compared to estimation-based method. (c) Anomalies map.