| China is the world's largest producer and producer of tea,and a large number of tea plantations bring economic benefits.Tea garden is an important part of agricultural ecosystem,but its monitoring is not as good as farmland,so it is an important basis for agricultural production,soil management decision-making and ecological construction of tea garden to obtain the spatial distribution information of tea garden in time and accurately.The planting of tea garden has changed the surface spectral characteristics and spatial structure of soil,so it is difficult to accurately identify tea garden only by the reflection spectral characteristics of ground features.The high-resolution remote sensing image data can capture the spatial detail information.For the objects with obvious texture structure such as tea garden,texture features can effectively improve the recognition accuracy,but the current texture extraction algorithm has strong pertinence and is difficult to satisfy the visual separability.Identification requirements for poor,diverse texture types,blurred boundaries,or complex texture images.Although machine learning and deep learning methods have made a breakthrough in the accuracy of object recognition and classification,but the method is complex,time-consuming,and requires a large number of training samples,under this background,the trained classifier often does not have portability.In order to solve these problems,this pap takes that planting area of Lancang tea garden in Yunnan province as an example and adopts a texture pattern extraction and enhancement algorithm of high-resolution remote sensing image to realize the object-oriented non-parametric tea garden identification method and realize fast and high-precision tea garden identification with minimum cost No,this study can be summarized into the following three aspects:In that first part,the tea garden texture enhancement algorithm base on spatial point pattern analysis is realized.The spatial texture features of Worldview-2 image data can compensate for the limitations of spectral features and improve the accuracy of tea garden identification.however,for complex texture images with diversified texture types,only the texture features can not accurately distinguish the features withdifferent texture structures In this paper,a spatial point pattern analysis method based on LBP texture feature is proposed to explore the spatial distribution pattern(ICS)of texture structure.The results showed that the tea gardens in the study area presented an aggregation distribution pattern,and the other categories presented an even distribution pattern or a mixed distribution pattern.Using separability to evaluate LBP texture feature and ICS feature to distinguish tea garden from background,the average distance between tea garden and background feature of ICS feature is 0.17,which is 0.11 more than LBP texture extraction result.Therefore,spatial point pattern analysis method based on texture feature image can be used To depict the details of tea garden,and to increase the differentiation between tea garden and other kinds of land.The object-oriented non-parametric tea garden identification method based on texture pattern was realized.In order to solve this problem,the computing cost of machine learning recognition method is high and the learning process requires a large number of training samples.In order to solve this problem,a non-parametric texture intensity tea garden recognition method based on the difference of spatial distribution patterns of different texture structures in texture images was proposed in this paper.No way.In that method,the concentric double window was set to detect the characteristic value of the uniform or mixed distribution pattern in the ICS characteristic,and the characteristic set of the aggregate distribution was retained.That is,the tea garden,the result shows that when the inner window of the concentric double window is set to 3 and the outer window is set to 17,The identification accuracy is the best: The overall accuracy is 92.90% and the Kappa coefficient is 0.81,which realizes fast and high-precision identification of tea garden.By analyzing the applicability of different feature extraction algorithms to tea garden recognition,an object-oriented feature combination tea garden recognition method based on machine learning is proposed,and the characteristics and application scope of different recognition methods were compared.In object-oriented tea garden recognition,different algorithms have different contributions to improving tea garden recognition accuracy.Compared with the commonly used GLCM,PSI,LBP texture feature extraction algorithms,the LBP texture feature combined withspectral feature has higher recognition accuracy: The overall accuracy is 91.68 %,Kappa coefficient is 0.81.At the same time,the applicability of ICS feature in machine learning and recognition method was discussed.The results show that the overall accuracy of the spectrum feature recognition of ICS feature combination is2.18 percentage points higher than that of LBP feature recognition,and the Kappa coefficient is increased by 0.05.Aiming at the problem of poor visual separability,fuzzy texture image and single texture feature extraction method,this paper uses GLCM,PSI,LBP and ICS to explore the contribution of feature combination to target object recognition,and the results show that the spectrum is special.When the characteristics of GLCM,PSI,LBP and ICS were combined,the identification accuracy is the best,the overall accuracy is 96.89%,the Kappa coefficient is 0.93,and the precise and high-precision identification of tea garden is realized.Compared with the two recognition methods,the precision of machine learning recognition is better than that of texture intensity recognition(the overall precision is 3.99 percentage points and the Kappa coefficient is 0.12)In this case,the texture intensity recognition algorithm proposed in this paper is more suitable. |
PDF Full Text Request