RESUMO
Taking 51 field measured hyperspectral data with different pest levels in Yanping, Fujian Province as objects, the spectral reflectance and first derivative features of 4 levels of healthy, mild, moderate and severe insect pest were analyzed. On the basis of 7 detecting parameters construction, the pest level detecting models were built. The results showed that (1) the spectral reflectance of Pinus massoniana with pests were significantly lower than that of healthy state, and the higher the pest level, the lower the reflectance; (2) with the increase in pest level, the spectral reflectance curves' "green peak" and "red valley" of Pinus massoniana gradually disappeared, and the red edge was leveleds (3) the pest led to spectral "green peak" red shift, red edge position blue shift, but the changes in "red valley" and near-infrared position were complicated; (4) CARI, RES, REA and REDVI were highly relevant to pest levels, and the correlations between REP, RERVI, RENDVI and pest level were weak; (5) the multiple linear regression model with the variables of the 7 detection parameters could effectively detect the pest levels of Dendrolimus punctatus Walker, with both the estimation rate and accuracy above 0.85.
Assuntos
Insetos/crescimento & desenvolvimento , Pinus/química , Pinus/parasitologia , Análise Espectral/métodos , Animais , Clorofila/análise , Modelos Lineares , Doenças das Plantas/parasitologia , Folhas de Planta/químicaRESUMO
Taking the images of Landsat TM, ALOS AVNIR-2, CBERS-02B CCD and HJ-1 CCD as the experimental data, for increasing the differences among shaded area, bright area and water further, the present paper construed a novel vegetation index-Shaded Vegetation Index(SVI), which can not only keep the absolute differences among bright area, shaded area and water area in the near-infrared band, but also can enlarge NDVI, eliminate the possible mixes, and change the histogram "skewed" phenomenon of NDVI, so the vegetation index value is closer to normal distribution, and more in line with the filed condition; this new index was applied to the surface features of large difference of the near-infrared radiation characteristics. Verified by accuracy assessment for the bright area, shaded area and water area recognition effects with SVI, it was showed that the overall classification accuracies of these images were up to 98. 89%, 100%, 97.78% and 97.78% respectively, with the overall Kappa statistics of 0.9833, 1, 0.9667, and 0.966 7, indicating that SVI has excellent detection effects for bright area, shaded area and water area; the statistical comparison of sub-images between SVI and NDVI also illustrated the reliability and effectiveness of SVI, which can be applied in the shadow removal for remote sensing images.