Species Classification Of Temperate Plantation Using Airborne Hyperspectral Images

Airborne hyperspectral remote sensing is a passive method of remote sensing with high spatial and spectral resolution,which can describe the texture and spectral information of features better.In this study,the main tree species of Mengjiagang Forest Farm were identified using AISA Eagle Ⅱ hyperspectral imagery with the spectral range of 400-1000 nm.In the classification,the vegetation index reflecting the difference of various pigment contents of vegetations and the texture factor reflecting the canopy structure were used together.The Minimum Noise Fraction(MNF)transformation method and the Recursive Feature Elimination(RFE)method were used to reduce the dimension of hyperspectral image.The extracted features were combined to explore the influence of different features on the classification accuracy of the tree species.The flight height is lower than the cloud layer,therefore the hyperspectral image is easily affected by the cloud layer during the data acquiring,especially in the forest area with the weather changing more rapidly.The shadow area and the non-shadow area images were classified separately in the classification.The main findings and conclusions of this study include:(1)For the two methods of extracting texture features,when verified the results using field data,the overall accuracy based on texture feature transformation is 97%,which is 1% higher than that based on spectral transformation.When verified the results with forest map,the overall accuracy and Kappa coefficient of the classification results based on texture feature transformation were 91% and 0.87 respectively,which were 2% and 0.02% higher than those based on spectral transformation.It shows that the texture information extracted by texture feature transformation is more effective for improving the accuracy of tree species recognition.(2)For non-shaded areas,the classification accuracy of MNF combined with vegetation index and texture information is higher than the accuracy of RFE combined with vegetation index and texture information when using filed data and forest map to verify.It shows that the MNF transformed images are more suitable for the classification of four conifers.For texture information and vegetation index,the classification accuracy of texture information combined with two dimensionality reduction methods was higher than the accuracy of vegetation index combined with two dimensionality reduction methods.It indicates that texture information has greater potential for improving four conifer species.(3)For the shaded area,the classification accuracy of MNF combined with vegetation index and texture feature is higher than that of RFE method and vegetation index and texture feature combination.It is indicated that MNF transformation method is more suitable for shaded image classification.When using the forest map to verify the results,the overall accuracy of the texture information combined with the two corresponding dimensionality reduction methods is higher than the accuracy of vegetation index combined with two dimensionality reduction methods.It shows that for the shaded area,the texture information contributes more to improve the classification accuracy of tree species than the vegetation index.(4)When the forest map was used to verify the classification results of the whole forest farm,the overall accuracy of the classification was 87.6%,indicating that the airborne hyperspectral image has great application potential in tree species classification.(5)By analyzing the classification accuracy of shaded and non-shaded areas,it can be seen that the combination of MNF transform and texture information had the highest classification accuracy.It indicates that same features can be chosen in the classification of non-shaded and shaded areas and then these two kinds of areas should be classified separately.