Dimensionality Reduction using Hybrid Algorithms and Their Application to Remote Sensing Data

• Main Authors: Maha Hasso, Mona Siddiq
• Format: Article
• Language: Arabic
• Published: Mosul University 2013-03-01
• Series: Al-Rafidain Journal of Computer Sciences and Mathematics
• Subjects: remote sensing data; k-mean algorithm; signal; noise ratio
• Online Access: https://csmj.mosuljournals.com/article_163463_5ef657044e6e091ef13ba7d02a8b3224.pdf

In this work, A proposed Algorithm has been constructed for the selecting the best band and lessening high dimension of remote sensing data depending on multi algorithms, each on carried out and its results are studied irrespective of other, then combining them in the proposed algorithms, in the principle component analysis algorithm find covariance matrix for the processing bands . Then find Eigen vector by using Jacobs’s method and this represents the highest value in Eigen vector. The algorithm was applied on many groups of multispectral image for the Mapper sensor. By applying it on the first group of images it concluded that the sixth band is the best one, because the value of its Eigen vector is the biggest one. when the algorithm was applied on the second group of images it concluded that the second band is the best one, and the value of its Eigen vector is the biggest one, when the algorithm was applied on the third group of images it concluded that the fifth band is the best, and the value of its Eigen vector is the biggest one (regarding separating the sixth infrared band in the three groups By using wavelet transform algorithm for one level of analysis and selecting the best band according to the least value of mean square error , to show the result of selecting the best ,the k_means algorithm  was used to classify images By using K_mean classification algorithm in images .A new way was proposed to determine  centers which is an important matter in accurate  classifications and specifying initial centers  by finding the maximum value and minimum value   and finding the mean between them until getting the wanted number of centers. Thealgorithm was applied on three groups of multispectral images .the classification was done on total number of bands to product one band out of it. A new algorithm was constructed depending on the previous three algorithms which applies the wavelet transform on multispectral images and finding the signal to noise ratio depending on variance of each band And arranging it decendingly and then choosing the bands that have highest datas in order to select the best bands and apply the principle component analysis on it. After finding the Eigen vector from the algorithm and selecting the highest values from it, it will be classified. From applying the proposed algorithm it has been clear that it is the best in accordance to applying .because it has shown high efficiency and accuracy in classification and in finding the best band.