Informations et ressources scientifiques
sur le développement des zones arides et semi-arides

Titre : Research on the Remote Sensing Image Classification Model and Spatiotemporal Change Pattern of Salinized Land in Arid Oasis Area ——The Example of the Taolai River Basin

Auteur : 刘宇航

Grade : Master 2022

Université : Lanzhou University of Technology

Résumé partiel
Soil salinization is the main obstacle restricting the efficient utilization of water and soil resources,sustainable development of ecological environment and food security in the arid oasis area of Northwest China.Conservation,efficient use of land and water resources,and food security are critical.Therefore,this paper selects the saline-alkali land in the oasis area of the Taolai River Basin,a typical arid oasis area,as the research object,and uses the long-term remote sensing images of this area from1990 to 2020.The classification methods of remote sensing images were compared to verify the reliability of the model,and the task of classifying remote sensing images of saline-alkali land in the region was completed,and the distribution map of salinealkali land was produced.And on this basis,the study and analysis of the temporal and spatial change pattern of saline-alkali land in the oasis area of the Taolai River Basin,and the dynamic change trend of the saline-alkali land in this area were ascertained.The main contents are as follows :（1）Taking the oasis area of the Taolai River Basin as the research area,through field investigation and field sampling,a remote sensing classification system was formulated in combination with the actual types of ground objects and the principles of ground object classification.According to the best band combination results,the remote sensing image interpretation mark is determined,and on this basis,the production of the remote sensing classification sample data set is completed.The remote sensing feature classification system and the corresponding number of samples in the study area are as follows : 1260 vegetation samples,1137 construction land samples,643 water samples,328 severe saline-alkali land samples,196 moderately saline-alkali land samples,and 235 moderately saline-alkali land samples.There were421 bare ground samples,and the sample separability was greater than 1.9,indicating good separability.（2）Based on the deep learning ENVI-Net 5 model architecture,the deep learning classification model of saline-alkali land was constructed by using the ENVI software Deeplearning module.According to the experimental results,the model is reversely optimized,and it is verified and compared with the traditional classification method to evaluate the accuracy of the model.The results show that the overall accuracy,user accuracy and Kappa coefficient of the ENVI-Net 5 classification model are better than the traditional classification methods,and the SVM accuracy is better than the RF and neural network among the three traditional classification methods.

Mots clés : Salinized land in arid oasis area ;Remote sensing image classification ;Deep learning algorithm ;Spatiotemporal change pattern ;Taolai River Basin ;

Présentation (CNKI)