This sub-theme aims to enhance the developing hyperspectral techniques and analyses. Hyperspectral imaging systems collect data both spatially and in a large number of spectral bands. The data collected can be processed to exploit both the spatial information and the spectral component, leading to improvements in the identification and discrimination of targets within a scene. Hyperspectral measurement techniques, simulation/predictive capabilities and processing algorithms are undergoing development. However, the true potential of hyperspectral sensing has yet to be exploited.


The benefit of hyperspectral imaging lies in the area of improvement in target detection and recognition within camouflaged, concealed or decoy-employed environments. This function can supplement existing detector technology, giving rise to higher probabilities of detection, whilst reducing the requirement for human


Review of current capabilities (measurement, simulation and algorithms) to highlight strengths and deficiencies
Analysis of typical hyperspectral imagery, leading to the development of more realistic hyperspectral scene simulation methods
Algorithm development, capable of adapting to various imaging conditions and implementation for real-time systems
Comparison of conventional and hyperspectral target detection techniques
Comparison of real and simulation-based detection capabilities
Recommendations on efficacy of hyperspectral bands under various criteria


Hyperspectral (and multi-sensor) systems make use of the systems’ dimensionality to improve target detection and recognition. This sub-theme will seek to exploit existing and developing algorithms using real and simulated imagery to enhance current target detection and classification performance.

All projects that will be undertaken initially in this sub-theme are closely linked, and aim to improve the technology readiness level from conceptual maturity through to applicable systems. Several required components have been highlighted to enable these improvements, starting with a review of current capabilities. By highlighting both strengths and deficiencies in current measurement, simulation and algorithm capabilities, the problem can be scoped and key development needs addressed.

Measured data are available, e.g. MUST 2000 trials, but obtaining high quality measured data on advanced systems in appropriate contexts is often problematic. Measured data are also limited to the context in which the data were gathered, including location, target/scene type, atmospheric conditions and sensor characteristics. It is therefore desirable to provide high quality simulated data that can be tailored easily to the specific needs of any hyperspectral assessment. Key components required for simulation will be highlighted and deficiencies in existing capabilities addressed.

Hyperspectral systems aim to improve threat detection and identification, by using the spectral and spatial dimensionality of the data gathered. Sophisticated algorithms are required to do this, and are primarily based on statistical pattern recognition techniques. Further development would focus on adaptive algorithms, enabling variations in measurement conditions to be accommodated.

This would include data reduction, selecting spatial and spectral subsets specific to the threat of interest. The ultimate aim would be to assess measured scenes in real-time, to determine the threat from difficult targets.

It is important to understand how far the spectral dimension can be utilised in improving target detection and recognition. Hyperspectral techniques will therefore be compared with conventional broad/single band capabilities. This will help indicate where improvements can be realised using hyperspectral as well as highlighting which bands are key and under what conditions.


Much research has already been conducted on aspects of hyperspectral imaging. BAE SYSTEMS, under the FOAS Improved Targeting TDP is analysing this technology as a potential sensor for long range target discrimination. This has involved generating sample simulated imagery, and comparison with available measured imagery. Simple anomaly detection algorithms have been applied to these data, incorporating band selection techniques. THALES and QinetiQ have been involved with the construction of a database and manufacture of an instrument for trials data gathering.

Leave a Reply

Your email address will not be published. Required fields are marked *