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

Accueil du site → Master → Chine → 2022 → Automatic Detection of Center Pivot Irrigation Systems in Africa with Sentinel-2 Data

University of Electronic Science and Technology of China (2022)

Automatic Detection of Center Pivot Irrigation Systems in Africa with Sentinel-2 Data

范涛

Titre : Automatic Detection of Center Pivot Irrigation Systems in Africa with Sentinel-2 Data

Auteur : 范涛

Grade : Master 2022

Université : University of Electronic Science and Technology of China

Résumé
Agriculture is the main source of economy for most African countries at present.Because Africa is located near the equator,the sunshine time is longer,and the precipitation distribution is less and very uneven.,resulting in many arid and semi-arid regions in Africa.Due to physical water shortage and relatively little precipitation,farmers generally need to adopt irrigation methods to increase crop yields while relying on rainfall.Irrigation is more effective than rainfall in enhancing and increasing crop yields.Center Pivot Irrigation Systems(CPIS)is a typical irrigation equipment and is widely used in the world.The center pivot irrigation system has a distinct shape,generally in the shape of a circle or a circular segment shapes.Through the center pivot irrigation system,we can understand the agricultural development status in Africa,and then estimate the agricultural output in Africa,which can provide a certain basis for resource allocation and agricultural planning in Africa,and can also provide assistance to countries in the arid and semi-arid regions of Africa for countries around the world.Data support is provided.Therefore,mapping Africa’s irrigated areas is very important for African agricultural management and water resources management.Due to the unique shape of the center-pivot sprinkler irrigation system,this paper uses deep learning to perform automatic detection and segmentation of the center-pivot sprinkler irrigation system.In this paper,the arid and semi-arid regions of Africa are used as the research area,and the Sentinel 2 image is used as the data source.First,the classical deep learning network is used to detect and segment the target,and the results are compared and analyzed,and finally the one with better effect is selected.Cascade R-CNN serves as the base network for this paper.On this basis,according to the characteristics of optical remote sensing images and center pivot irrigation system,corresponding improvements are made on the network to improve the performance of the model.For the backbone network,based on Res Net101,this paper introduces the Convolutional Block Attention Module(CBAM)and improves it,so that the features extracted by training can better focus on the focus channel and spatial information.For the neck network,this paper introduces the feature pyramid structure,and on this basis,designs a semantic enhancement module and an attention weighting module.This enables the network to obtain richer contextual information,optimizes the contribution of feature maps at each level during feature fusion,and obtains a better feature pyramid model.The resulting model improves m AP by nearly 3 percentage points compared to the original model.And use the algorithm to conduct a comprehensive inspection of the arid and semi-arid regions of Africa,draw the first 10 m resolution distribution map of the center pivot irrigation system in Africa,and count the area of the center pivot irrigation system in various African countries.This will help to better understand the current water use situation in Africa,help countries in arid and semi-arid regions of Africa to make a clearer judgment on the potential of their agricultural modernization,and further develop better agricultural management and water management policies

Mots clés : deep learning ;center pivot irrigation system ;feature pyramid network ;cascade r-cnn ;CBAM;

Présentation (CNKI)

Page publiée le 24 octobre 2022