Band selection for Japanese oak wilt extraction in autumnal tints of forest based on NWI

Extraction of Japanese oak wilt in autumnal tints of forest based on remote sensing data is required from the view point of early detection and prevention, because the appearance of oak wilt ranges from early June to October. In previous work, we have proposed an oak wilt index, NWI, in which extraction accuracy from a wide variety of autumnal tints has not been verified. At first, it was confirmed that NWI is not capable of distinguishing wilt regions from healthy late stage tints. An additional band, 515nm, to NWI-related three bands, i.e. 800, 680 and 550nm, are selected by SVM-based analysis evaluating the ratio of support vectors and the contributions in four dimensional normalized reflectance space spanned by NWI-related three bands and the candidate band. We proposed an index, NDIwilt, to differentiate wilt from late stage tints in the framework of normalized vegetation indices based on two bands. By the combination of NWI and NDIwilt, oak wilt regions are extracted with high accuracy, Az value 0.9999.