Hyperspectral anomaly detection of hidden camouflage objects based on convolutional autoencoder

Kuester, Jannick; Gross, Wolfgang L.; Middelmann, Wolfgang

doi:10.1117/12.2598011

Cited by 1 publication

(1 citation statement)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Hence, one basis behind the anomaly detection issue is the identification of desired targets that are unknown in advance and whose existence could be indicative of a suspicious behaviour. This lack of previous knowledge turns the detection of anomalous spectra into an essential matter in military and civilian applications, such as defense and surveillance, environmental monitoring, rare natural disaster detection, agriculture studies, among others [2], [3].…”

Section: Introductionmentioning

confidence: 99%

Low-Power Hyperspectral Anomaly Detector Implementation in Cost-Optimized FPGA Devices

Caba

Díaz²,

Barba

et al. 2022

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

Onboard data processing for on-the-fly decisionmaking applications has recently gained momentum in the field of remote sensing. In this context, hyperspectral anomaly detection has received a special attention since its main purpose lays on the identification of abnormal events in an unsupervised manner. Nevertheless, onboard real-time hyperspectral image processing still poses several challenges before becoming a reality. This is why there is an emerging trend towards the development of hardware-friendly algorithmic solutions embedded in reconfigurable devices. In this context, this work contributes with a hardware architecture that ensures a progressive line processing in time-sensitive applications limited by the scarcity of hardware resources. In this sense, we have implemented the state-of-theart HW-LbL-FAD detector on a reconfigurable hardware for a real-time performance. Specifically, we have selected a costoptimized FPGA (ZC7Z020-CLG484) to implement our solution whose results draw up a good trade-off between the following three features: time performance, energy consumption and cost. The experimental results indicate our hardware component is able to process hyperspectral images of 825x1024 pixels and 160 bands in 0.51 seconds with a power-budget of 1.3 watts and a device cost around 150 C. Regarding detection performance, the HW-LbL-FAD algorithm outperforms other state-of-the-art algorithms.

show abstract

Section: Introductionmentioning

confidence: 99%