Research On Rice Area Extraction Method Based On UAV Remote Sensing Technology

Rice is one of the world 's major food crops and a staple food for 70% of China 's population.If rice classification information can be obtained quickly and accurately,it can not only monitor the growth of rice in real time,but also provide timely and accurate agricultural information,which is of great significance for formulating agricultural policies and promoting the development of precision agriculture in China.Nowadays,with the popularization of high-resolution remote sensing images,the resolution of the images used in object-oriented ground feature extraction is getting higher and higher,Although many scholars have proposed improved remote sensing classification methods to obtain feature information in high-resolution images,there are few studies on rice extraction in high-resolution remote sensing images of domestic drones.Therefore,the rice information extraction of high-resolution remote sensing images of unmanned aerial vehicles still needs further study.This paper conducts an in-depth study on rice information extraction methods through high-resolution remote sensing image data obtained by the UAV in the field.First,the object in the image data is segmented and characterized under object-oriented segmentation technology,and then a feature classification system is established,and the threshold of classification rules is determined by introducing the concept of confidence interval of statistical theory and expert prior knowledge.Finally,a multi-level object classification model is constructed based on the object-oriented rule set classification method,and the model is used to extract rice information.The specific research content and results are:(1)According to the actual cultivation of rice in the study area,the growth characteristics of rice were investigated,and the image characteristics information of rice,other ground features such as spectrum,texture,and geometry were analyzed in detail by UAV remote sensing images,and a classification system of ground features was established.(2)The object-oriented segmentation technology is used to conduct image segmentation experiments on various types of objects in UAV remote sensing images,and analysis of several common image segmentation methods(including: checkerboard segmentation,quadtree segmentation,spectral difference segmentation,and FNEA)Multi-scale segmentation method),the results are obtained through experimental comparison: FNEA multi-scale segmentation method has better ability to segment the boundary of high-resolution image features,and optimizes the segmentation scale and segmentation parameters of various features.(3)The rule-based classification method and several commonly used sample-based classification methods(support vector machine,Bayes)in object-oriented classification are studied.Concept of confidence interval and prior knowledge of experts to determine the threshold of classification rules,and a multi-level classification model for rice extraction is proposed.The accuracy of the classification results is evaluated and analyzed by the accuracy test indicators of production accuracy,user accuracy,total accuracy and Kappa coefficient in the confusion matrix.The experimental results show that the multi-level feature classification model combined with the object-oriented rule set classification method has a good effect in the extraction of rice information,especially in the case of comparing training based on small samples,the classification accuracy of the model is significantly better than the classification accuracy based on single rule set classification and sample-supervised learning.(4)Extract the rice area,calculate the rice area extracted by various methods through Arcgis software,and the statistical interpretation of the rice area extracted by visual interpretation is used to evaluate the accuracy and suitability of various classification methods for extracting rice area from high-resolution UAV remote sensing images.