Selection of optimum bands combination based on multispectral images of UAV

With the rapid development of unmanned aerial vehicle (UAV), it is widely used in the field remote sensing which is different from the satellite remote sensing, and has many advantages such as more convenient, lower cost, and shorter revisit cycle. However, effective information cannot be extracted from the multispectral images of UAV easily because of the high-resolution multi-band redundant data which can increase the complexity of data processing and consume a lot of computational resources. Therefore, the purpose of this research is to study the optimum bands combination which can be extracted by multispectral image. Manas's riverside in Shihezi, Xinjiang was selected as research area. Fixed-wing UAV equipped with Micro MCA12 Snap was used to obtain high-resolution multispectral images. Based on this system, a method was proposed to select the optimum bands combination for topographical objects classification. First, the standard deviation and correlation coefficient of the multi-spectral image's gray value were analyzed; the original bands combinations were got with the OIF method. Then, the most informative spectral feature bands and texture feature bands were determined respectively by using variety methods, such as vegetation and water index, principal component analysis, and GLCM. Finally, the original bands combination, spectral feature bands and texture feature bands were combined to obtain the final result. According to the analysis, bands 1, 6, 11, NDVI, NDWI and the mean parameter of GLCM combination of Micro MCA12 Snap multi-spectral sensors were selected as the optimum bands combination for topographical objects classification. After the selection of the bands combination, unsupervised classification and supervised classification methods were used to verify the classification accuracy with the optimum bands combination respectively. The classification accuracy with IsoData of ROI (region of interest) was increased from 83.57% to 89.80%, when it comes to SVM, the accuracy was increased from 95.58% to 99.76%. In addition, the study also provides effective reference for the selection of optimum bands combination with UAV multispectral images. © 2016, Chinese Society of Agricultural Machinery. All right reserved.