The Forest Species Recognition System Based On Timing Optimization Fitting

The forest monitoring based on remote sensing image is highly valued,but the diversity and complexity of the forest have made many difficulties in the research of forest species recognition.In recent years,with the accumulation of remote sensing data,the corresponding phenological curves were obtained by timing fitting of forest's remote sensing time series data.The difference among phenology of different forest species was used to carry out forest species identification research.That has become a new research focus.In parts of the subtropical monsoon climate,the time of forest phenotypes changed obviously in spring and autumn is often cloudy or rainy weather.The natural conditions such as rainy will cause serious noise to the remote sensing images,which will cause the reduction of the high-quality remote sensing data and the lack of remote sensing data in critical period.The existing fitting and classification methods based on medium resolution remote sensing data have some problems such as: the phenological fitting curve is not accurate enough,the applicability of the classification method is poor and the recognition rate of forest species is low.In this paper,we propose a remote sensing data fitting method based on sparse and effective time series data,which can improve the accuracy of phenotypic curve fitting of forest species under the condition of less human intervention.Then a method of forest classification recognition based on medium precision remote sensing data is proposed by combining phenological characteristics,spectral characteristics and texture features of the forest,which achieve a more accurate classification than other methods.The main contents of the thesis are as follows:(1)In order to obtain the accurate forest phenology curve,a sparse effective time series remote sensing data fitting method is proposed and designed.The denoised remote sensing time series data are fitted based on the second-order Gaussian,then high quality vegetation phenology curve is carried out through weighted iteration by adding the weight estimation of the fitting result residual,which can be better used for extracting phenology feature.Through the experiment,compared with other methods,the method proposed in this paper can improve the fitting precision effectively.(2)In order to distinguish the forest species with similar phenotypes more accurately,this paper proposes a method of forest species classification based on medium precision remote sensing data with the phenological characteristics,spectral characteristics and texture features of different forest species.First of all,through the analysis in the separability of forest characteristics,the three types of features were effectively screened and dimensioned.And then preprocess the data feature set.Finally,SVM classification method is used to realize the training and classification of forest species.The results show that the method is more accurate than other commonly classification methods.(3)Finally,this paper designs and implements the forest recognition system based on the above two methods.The system is able to achieve forest classification included four modules: remote sensing data preprocessing,timing fitting,feature extraction and forest species classification.The user can do some simple operations through the interactive interface to obtain the timing fitting results and classification results.This system not only provides a better user physical examination,but also further verify the practicality of the method.