Multi-spectral Remote Sensing Recognition On Main Deciduous Species In Mount Tai

The accurate recognition on mountainous tree species is the key and difficult point of remote sensing recognition,and the recognition accuracy directly affects the accuracy of forest remote sensing mapping.In recent years,remote sensing technology has been widely used in forest tree species recognition,which provides an important reference for precise recognition of mountain tree species.Multi-spectral remote sensing tree species recognition has the characteristics of macroscopical,short cycle,repeatable and so on,which has advantages in large spatial scale survey.However,the band value was always as spectral information directly used in the multi-spectral remote sensing tree recognition,and the spectral index can enlarge the small difference between the spectrum and improve the recognition accuracy of the tree species.In the existing methods of tree species recognition,support vector machines are widely used and have high precision,and the cloud model recognition is effective,but there are few researches on tree species recognition by cloud model.Identification of the optimum phase,recognition model,spatial feature information is particularly important to improve the recognition accuracy of the deciduous broadleaf tree species in Mount Tai.Multispectral images of ZY-1 02C and ZY-3 at three phases were selected to optimize sensitive spectral indices which were introduced into support vector machine to choose the optimum phase of recognizing on Quercus acutissima and Robinia pseudoacacia.The sensitive bands were selected,sensitive spectral indices and texture parameters were constructed and screened based on the spectral and texture features from multispectral remote sensing images of ZY-1 02C and ZY-3 at three phases（May 12,September 29th and December 7th）,then they were put into cloud model,support vector machine,and maximum likelihood classification model for the remote sensing recognition on main deciduous species in Mount Tai（a case study with Quercus acutissima and Robinia pseudoacacia）.The best model,the optimum phase and the best characteristic information for tree species recognition were obtained,and the spatial distribution of Quercus acutissima and Robinia pseudoacacia in Mount Tai was retrieved based on the optimal recognition model.The main conclusions of the study include:（1）Spectral characteristics and sensitive spectral indices were clarified.In terms of spectral characteristics,the average spectral reflectance of Quercus acutissima was higher than that of Robinia pseudoacacia,especially B₄ on May 12th.The sensitivity bands of all phases were consistent,concentrated on B₄ and B₃,among which B₄ on May 12th was the most sensitive band.Correlations between spectral index and tree species of May 12th was higher than those of December 7th,and they both higher than September 29th,and the spectral indices constructed by band 3 and 4 were higher than other bands at all three phases.（2）The optimum phase of the tree recognition was determined.One-dimension cloud model had the highest recognition accuracy on all three time phases,the overall recognition accuracy on the two species was 91.86%on May 12th（94.18%on Quercus acutissima,89.54%on Robinia pseudoacacia）,79.16%on September 29th（87.29%on Quercus acutissima,71.02%on Robinia pseudoacacia）,90.57%on December7th（91.65%on Quercus acutissima,89.49%on Robinia pseudoacacia）.The recognition accuracy on May 12th was slightly higher than that of December 7th,far higher than September 29th,and May 12th was the optimum phase of the tree recognition.（3）The best model and the optimal spatial characteristics of the tree recognition was obta ined.May 12th had the highest recognition accuracy by all three models,the overall recognition accuracy on the two species was 91.86%by cloud model（94.18%on Quercus acutissima,89.54%on Robinia pseudoacacia）,89.25%by support vector machine（93.53%on Quercus acutissima,84.98%on Robinia pseudoacacia）,81.06%by maximum likelihood classification model（86.27%on Quercus acutissima,76.35%on Robinia pseudoacacia）.The recognition accuracy of cloud model was slightly higher than that of support vector machine,far higher than maximum likelihood classification model,and cloud model was the best model of the tree recognition.May 12th with cloud model had the highest recognition accuracy based both on spectral and texture features,the overall recognition accuracy on the two species was 91.86%based on spectral features（94.18%on Quercus acutissima,89.54%on Robinia pseudoacacia）,79.45%based on texture features（86.43%on Quercus acutissima,72.46%on Robinia pseudoacacia）,the recognition accuracy based on spectral feature was far higher than that of texture feature.（4）The spatial distribution of Quercus acutissima and Robinia pseudoacacia in Mount Tai was retrieved.The spatial distribution of Quercus acutissima and Robinia pseudoacacia in Mount Tai was retrieved based on the cloud model with spectral feature of May 12th.The user precision of Quercus acutissima was 92.00%,and of Robinia pseudoacacia was 89.00%,the overall recognition accuracy on the two species was 90.50%,with kappa coefficient was 0.8946.The distribution area of Quercus acutissima and Robinia pseudoacacia were 15.22 km² and 7.83km²,and the Quercus acutissima was distributed in the Midwest and the north of Mount Tai,the Robinia pseudoacacia was spread in the river side,foothills,mountainside and mountain slope of central Mount Tai.The result showed that the recognition accuracy of the model based on sensitive spectral index was obviously higher than that based on sensitive band,especially for Robinia pseudoacacia.The recognition accuracies on Quercus acutissima was higher than Robinia pseudoacacia.The recognition results of three-dimensional cloud model were consistent with the one-dimensional cloud model both,and the accuracy of three-dimensional cloud model was close to the highest accuracy of the one-dimensional cloud model.The one-dimensional cloud model was simple and less calculation,so it can be used as the priority cloud model.In summary,the cloud model constructed by the sensitive spectral index based on the spectral information of May 12th had the highest recognition accuracy on Quercus acutissima and Robinia pseudoacacia in Mount Tai.This study will provide a technical support for the precise recognition and remote sensing mapping of Mount Tai tree species.