The Application Of Minimum Noise Fraction That Based On Morphological Filter

There is a lot of terrain reflection spectrum information in the hyperspectral data.Because of the influence of the atmosphere, sensor, data transmission and so on, rawimage data is polluted largely. All of these disturb the absorption of reflectionspectrum, not only lower the precision, but also make it very inconvenient insubsequent processing. So, it is the indispensable key step of hyperspectral remotesensing image to analysis and eliminate the noise in the pretreatment. Meanwhile,abundant information is contained in the hyperspectral remote sensing image, butredundant information has no practical significance in the research and occupiessegmental storage space. It also affects the efficiency of the processing. Thus, weconsider to reduce the dimensions of the image data and this operation always haslittle influence on the subsequent processing and will reduce the amount of computingof data in a large extent. So, it is very necessary to achieve dimensionality reduction.The dimension reduction of hyperspectral remote sensing image has manymethods. According the results of dimension reduction, there are two types which arebased on feature extraction and band selection separate. Minimum Noise Fraction(MNF) is a dimension reduction method that based on feature extraction. It canquickly and efficiently compress hyperspectral image information to lower dimension.Minimum Noise Fraction, a linear transformation as a measure of MaximumSignal to Noise Ratio (MSINR), separates the noise and at the same time realizesdimension reduction. Essentially, Minimum Noise Fraction is the PrincipalComponent Analysis (PCA) which superimposed twice. Firstly, choose the high-passfilter template to deal with the whole image and then get the Noise covariance matrix.Secondly, to combined with the noise covariance matrix which was gotten on the firststep, complete the MNF dimension reduction process. This article chooses the hyperspectral remote sensing data which comes fromDexing copper mine of jiangxi province to do MNF. In the first step, chooseMathematical morphology method to construct the weighted multi-levelmorphological filter of full-range structural elements, then use gradient operator to getthe noise covariance matrix. Here we make a comparative analysis with the filteringeffect of an ideal high-pass filter. In the second step, make a principal componentanalysis on the covariance matrix of the original image. Through the above two steps,we finish the MNF. At this time, when we regard contribution rate as standard we canfind the MNF s dimension reduction effect which based on the full range of structuralelements weighted multi-level morphological is better than the original one. It can beshown that the effect will be better if Morphological filtering is applied to theminimum noise separation, and that the choice of structural elements influence thefinal dimension reduction effect.