Research On Extracting Citrus Orchard Based On Texture Features Of High Resolution Remote Sensing Images

Citrus is currently the fruit with the largest cultivated area,highest production,and highest consumption in China.The citrus industry in our country is well-developed,with a large number of practitioners and extensive planting areas.Therefore,the rapid and accurate acquisition of distribution information of citrus orchards from high-resolution images plays an important role in understanding the spatial distribution of citrus cultivation and real-time monitoring.This study focuses on the extraction of citrus orchards based on the texture features of high spatial resolution remote sensing images.The main research content and conclusions are as follows:(1)In this study,high-resolution GF-7 and GF-1 remote sensing images were used as data sources.Texture features were extracted using three methods: Gray-Level Cooccurrence Matrix(GLCM),Local Binary Patterns(LBP)operator,and Laws texture energy measurement.The texture features extracted by different methods were then combined with spectral features and vegetation index features to construct four different feature combination schemes.These schemes were used in classification experiments on remote sensing images using the random forest classification algorithm.The experimental results of different classification schemes were compared to analyze the texture methods suitable for extracting citrus orchards.Additionally,remote sensing images of different spatial scales were constructed in the study area to explore the spatial scale suitable for extracting citrus orchards through the recognition and extraction of citrus orchards in different resolution remote sensing images.(2)Based on GF-7 remote sensing images,extraction and classification experiments were conducted on citrus orchards in the study area,and the influence of different feature combinations on classification performance was investigated.The experimental results showed that the introduction of texture features effectively improved the classification and recognition performance.Compared with the scheme that only integrated spectral features and vegetation index features,the schemes with added texture features achieved certain improvements in classification accuracy.Among them,the scheme with the LBP operator texture feature had better recognition and extraction performance for citrus orchards,with extraction accuracy(Fa)and overall classification accuracy(OA)reaching 92.89% and94.34%,respectively.These values were 1.38% and 0.9% higher than the scheme with spectral + vegetation index + Laws texture feature,and 2.58% and 1.79% higher than the scheme with spectral + vegetation index + GLCM texture feature.(3)Citrus orchard classification and extraction experiments were conducted on images with four different spatial resolutions: 0.65 m,2m,4m,and 8m,using the GLCM,LBP operator,and Laws texture energy measurement methods.The results showed that the combination of LBP operator and 2m resolution achieved the highest extraction accuracy for citrus orchards,with an accuracy of 93.24% and an overall classification accuracy of94.87%.The research conclusions provide some reference value for the application of orchard remote sensing monitoring and agricultural production.