Remote Sensing Extraction Of Tea Garden Based On Object-oriented And Multi-source Data Fusion

Tea, one of the most popular drinks in the world, has a long history and culture in China. It integrates ecological benefit, economic benefit and social benefit as a whole. However, the widespread planting of tea garden has positive and negative effects on ecological environment security. Although it has the function of conserving water, carbon fixation and oxygen production, cleaning up the environment, since most tea gardens planted on hilly slopes by destroying grassland and woodland, tilling the soil, may result in the loss of soil and water and bring some ecological environment problems. Therefore, studying the accurate range of cultivation of tea has very important scientific and practical significance for coordinating the development of regional economy and ecological environment. This study introduced the image object and the multi-scale segmentation technology, described and illustrated the spectral features, geometry and texture features which are used to describe the image object. In addition, based on the data of GF-1 satellite images, applying the multi-level rules classification method based on object-oriented and multi-sources data fusion to the study of remote sensing extraction of tea plantations. The optimal segmentation scale selection was carried out for different objects in the multi-level classification rules.We established the knowledge rules for the extraction of tea plantations, analysised the spectral characteristics, texture features and terrain features of different land cover types by making full use of multi-sources data aided. At the same time, we selected three kinds of good performance classification algorithms:maximum likelihood method, support vector machine (SVM) and random forests, supervised classification was carried out on the basis of object-oriented, and compared to the results with the method of multi-level classification rules on the accuracy, conclusions are as follows:(1) Using the classification method of object-oriented and multi-sources data fusion of multi-level rules, we find the overall classification accuracy and kappa coefficient are 88.07% and 0.85 respectively. The overall classification accuracy improved by 17.67%,13.87% and 14.97% when compared to three kinds of supervised classification methods based on oriented-object (maximum likelihood, and support vector machine (SVM), random forests), the precision has been greatly improved. And, in complex topography areas, vegetation types easily confused circumstances, the supervised classification methods, based on machine learning, is not very good to meet the classification requirements.(2) Using the classification method of multi-level rules, the tea producer's accuracy and user's accuracy are 88.2% and 87.7% respectively. It increased by 28.6%, 6.5% and 17.4% respectively on the producer's accuracy when compared to the maximum likelihood, SVM, and the random forests methods, and increased by 6.5%, 12.2% and 9.2% respectively on the user's accuracy. Therefore, the classification method based on object-oriented and multi-source data fusion of multi-level rules has a good effect on the extraction of tea plantations and it can meet the requirements of analysis and application.(3) After analysis and comparison, the classification method of multi-level rules based on object-oriented has the following advantages:1) It is able to use different classification methods flexibly according to different stages and different land cover types, integrate advantages of each classification method; 2) It can make comprehensive use of multi-sources and multi-dimensional data, handle both continuous and discontinuous variables, use the multi-sources data to obtain more comprehensive knowledge of the rules; 3) For different objects, it is able to extract the target objects flexibly in different scales, and a variety of land cover types on the same scale layer, and it can also be choosed to extract gradually on the optimal segmentation scale so as to get the best results.The main innovations of this paper:(1) The classification method based on object-oriented and multi-sources data fusion of multi-level rules is applied to the remote sensing extraction of tea plantations, which provides an effective method for remote sensing monitoring of tea plantations.(2) In the fuzzy classification of vegetation and non-vegetation, the roughness texture based on Gdal is introduced, which has achieved good results and laid a foundation for the accurate extraction of the tea plantations.(3) The classification method based on object-oriented of multi-level rules makes full use of the auxiliary function of multi-sources data to construct the knowledge rules. In different stages of classification and classification of different ground objects, a variety of classification methods can complement each other, and to play their respective advantages.