基于衛(wèi)星遙感影像的茶園精準(zhǔn)識(shí)別研究

打開(kāi)文本圖片集

Research on Accurate Identification of Tea Gardens Based on Satellite Remote Sensing Images

Li Dongxue'，Li Feng2， Zhao Mengjie'，Ding Zhaotang3，Ma Qingping'（1. Collge of Agriculture and Biology， Liaocheng University， Liaocheng 2520oo， China;2. Shandong Climate Center， Jinan 250031，China ;3. Tea Research Institute， Shandong Academy of Agricultural Sciences， Jinan 25010o， China）

AbstractDue to the complex and varied textural features and spatial distribution characteristics of tea garden in satelite remote sensing images，it is a challenge to effectively recognize tea garden using traditional methods.Therefore，this study was aimed to achieve precise identification of tea gardens through satelite remote sensing images combined with high-precision semantic segmentation technology.Firstly，according to the satelite images of Rizhao City，the Labelme tool was utilized to accurately annotate the textural information of tea gardens and construct a dataset of tea garden remote sensing images.Based on this dataset，seven semantic segmentation models such as Vgg-unet， Resnet5O-unet and Segformer_b2 were trained using data augmentation techniques for precise identification of tea gardens.The results indicated that among the seven models，Segformer_b2 showed the best property with accuracy of 91.29% ，recall rate of 91.09% ，mean intersection over union（mIoU）of 84.45% and F1-score of 91.20% for tea garden remote sensing image segmentation. To further enhance the performance of the Segformer_b2 model in identifying tea gardens，the multi-head context attention（MCA）mechanism was introduced，and strided convolution was integrated to enhance the model's ability to capture key textural features of tea gardens and its perception and processing capabilities for local details in the images.The improved Segformer-b2 model exhibited enhanced performance in tea garden remote sensing image segmentation with accuracy increased to 91.31% ，recall rate increased to 92.97% ，mIoU increased to 85.87% ，and F1-score increased to 92.12% ，which could achieve the high precision and efficiency identification of tea gardens remote sensing images.This study could not only provide strong support for future precise identification and management of tea gardens，but also showcase the potential of satelite remote sensing technology in the agricultural field.

KeywordsSatellite remote sensing images； Tea garden recognization; Semantic segmentation model

遙感技術(shù)具有信息獲取迅速、觀測(cè)范圍廣等優(yōu)點(diǎn)，在資源普查、土地利用規(guī)劃、環(huán)境監(jiān)測(cè)等領(lǐng)域得到廣泛應(yīng)用[1-4]。（剩余9837字）