A partial convolution autoencoder that employs 1D convolution layers to capture local spectral features to handle the complex data distribution of normal samples, and employs the nearest-neighbor similarity score computed against normal features stored in the memory bank as the anomaly score.
Abstract
Hyperspectral imaging provides rich spectral information representing material-specific signatures that can be used as a non-invasive tool for detecting contaminants in the food industry. However, contaminant detection is challenging because their occurrence is rare and their variations are high. These characteristics make their detection well-suited to an anomaly detection framework in which samples are classified as either normal or anomalous. Existing deep learning approaches for hyperspectral anomaly detection commonly rely on reconstruction tasks, which are less effective when the normal pixels exhibit complex data distributions. Additionally, the use of fully connected layers and standard autoencoders for reconstruction does not align with the observation that spectral anomalies often manifest within specific wavelength ranges. To address these limitations, this study proposes a partial convolution autoencoder (PCAE) that employs 1D convolution layers to capture local spectral features. A two-phase training strategy is adopted: a self-supervised training phase to learn intrinsic spectral representations, followed by a one-class learning phase to obtain the compact features of normal data. To handle the complex data distribution of normal samples, during inference, this study employs the nearest-neighbor similarity score computed against normal features stored in the memory bank as the anomaly score. Evaluations on three hyperspectral food contaminant datasets showed that the proposed method achieved an average Area Under the Precision-Recall curve (AUPR) of 85.79%, outperforming current state-of-the-art methods while using 13.6 times fewer parameters. Further experiments on synthetically corrupted hyperspectral data showed that PCAE achieved the highest AUPR, demonstrating its robustness. These results demonstrate that PCAE is a promising model for food safety inspection and hyperspectral anomaly detection. Code will be released at https://github.com/Weejaa04/HSIAnomaly
A novel knowledge transfer paradigm from classification tasks to anomaly detection is proposed, which employs a pretrained convolutional enhanced Transformer (CET) classification network as the teacher model, transferring its rich spectral–spatial context encodings via feature-level distillation.
Lei Xia, Yin-Nian Liu· Remote Sensing· 0 citations
Hyperspectral anomaly detection aims to identify pixels that significantly deviate from the surrounding background without prior knowledge of anomaly classes. Existing reconstruction-based methods usually learn background representations through autoencoders and detect anomalies using reconstruction residuals. However,...
A self-adaptive background-attentive network with anomaly suppression (BASNet) is proposed, and a pyramid similarity downsampling module (PSDM) is employed to reduce the proportion of anomalies in the input image and improve multiscale background representation.
Xu Liu, Hao Yuan, Bo-Hui Tang et al.· IEEE Transactions on Neural...· 0 citations
Hyperspectral anomaly detection (HAD) aims to distinguish anomalies from background under an unsupervised paradigm and has been widely used in various fields. Recently, reconstruction-based deep learning methods have attracted considerable attention in HAD, but they have a limitation that anomalies tend to be incorpora...
Jie Xu, Lei Sun, Zi-Xi Yuan et al.· IEEE Journal of Selected Top...· 0 citations
Experiments show that HAD-MSF achieves consistently competitive detection performance with relatively low arithmetic complexity, demonstrating the effectiveness of integrating wavelet, state-space, and graph modeling for hyperspectral anomaly detection.
Jin-Zhuang Xu, Cheng-Long Zhang, Xiao-Xue Wang et al.· Neural Networks· 0 citations
Hyperspectral anomaly detection (HAD) is a highly challenging task, primarily because weak anomalous targets within high-dimensional spectral data are easily obscured by complex and heterogeneous backgrounds. However, existing methods still have limitations. They typically rely on strong background assumptions or focus...
Xiao-Bin Zhao, Yu-Xiao Wang, Jun-Jun Yin et al.· IEEE Transactions on Geoscie...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.