Research On Target Detection Methods For Full-band Hyperspectral Images

Target detection has always been a core technology in many military and civilian fields such as geographic information interpretation,military target detection,and regional security and fire prevention.This thesis closely combines the upsurge of full-band technology application that hyperspectral development is facing,and provides a set of technical solutions for detecting abnormal targets based on the hyperspectral data obtained in airborne application environment.Specifically,it includes band selection methods suitable for full-spectrum hyperspectral data,direct use of morphology and spectroscopy anomaly detection methods,and a hierarchical detection and identification spectrum matching method in combination with spectral library target reference spectra.The specific research contents are as follows:Chapter 2 introduces the band selection technique suitable for target detection.First,the sub-sets of the full-band hyperspectral data are divided by the mutual information between adjacent bands,and then the band selection based on the best index is performed within each sub-set.This scheme ensures that all band information can be embodied in the final selected band while reducing the amount of data redundancy.And it is very suitable for the later application of matching detection based on spectral library reference spectra.Chapter 3 introduces the method of abnormal target detection based on hyperspectral data.The most prominent feature of this type of definitive detection of suspected targets is the combination of the full band spectral information of the target,not just the morphological information.At present,the RX test is the most representative method in the field of anomaly detection of two-dimensional images,and the RX test represents a new type of test method for constant false alarm rate,which can be derived from a variety of extensions in morphology.With the ability to detect three-dimensional hyperspectral data.This chapter extends the RX algorithm to 3D hyperspectral data and gives experimental analysis.Chapter 4 proposes the overall scheme of hyperspectral image matching detection.In the scheme,the strategy of joint application of primary matching and secondary matching is planned.At the same time,it also refines the specific application scenarios of spectrum segment matching and process segment matching.A full-band hyperspectral initial matching method was proposed,and a special matching algorithm was screened and determined for the hyperspectral data of visible light,near-infrared and short-wave infrared,respectively,and the detection effect was better than that of the full-band overall matching.For the hyperspectral secondary matching method,the classification calculation based on the support vector machine is adopted,and the classification accuracy is improved significantly compared with the primary matching only.