Integrating Multitemporal Sentinel-1/2 Data for Coastal Land Cover Classification Using a Multibranch Convolutional Neural Network: A Case of the Yellow River Delta

Abstract

Coastal land cover classification is a significant yet challenging task in remote sensing because of the complex and fragmented nature of coastal landscapes. However, availability of multitemporal and multisensor remote sensing data provides opportunities to improve classification accuracy. Meanwhile, rapid development of deep learning has achieved astonishing results in computer vision tasks and has also been a popular topic in the field of remote sensing. Nevertheless, designing an effective and concise deep learning model for coastal land cover classification remains problematic. To tackle this issue, we propose a multibranch convolutional neural network (MBCNN) for the fusion of multitemporal and multisensor Sentinel data to improve coastal land cover classification accuracy. The proposed model leverages a series of deformable convolutional neural networks to extract representative features from a single-source dataset. Extracted features are aggregated through an adaptive feature fusion module to predict final land cover categories. Experimental results indicate that the proposed MBCNN shows good performance, with an overall accuracy of 93.78% and a Kappa coefficient of 0.9297. Inclusion of multitemporal data improves accuracy by an average of 6.85%, while multisensor data contributes to 3.24% of accuracy increase. Additionally, the featured fusion module in this study also increases accuracy by about 2% when compared with the feature-stacking method. Results demonstrate that the proposed method can effectively mine and fuse multitemporal and multisource Sentinel data, which improves coastal land cover classification accuracy.

Mapping Irrigated Areas Using Sentinel-1 Time Series in Catalonia, Spain

“…The S1 satellites (S1A and S1B) provide an exceptional combination of high spatial and high temporal resolutions for dual polarization SAR data (six days of temporal resolution and a 10 m × 10 m pixel spacing) available in free open access mode. S1 time series has been newly exploited by several studies for land cover classification [20,[31][32][33]. High quality classification mapping has been produced by applying either classical machine learning, such as random forest and the support vector machine, or complex deep learning methods on S1 time series.…”

“…High quality classification mapping has been produced by applying either classical machine learning, such as random forest and the support vector machine, or complex deep learning methods on S1 time series. Recently, deep learning (DP) techniques [32,34,35] have shown that neural network models are well adapted tools to automatically produce land cover mapping from information coming from both optical [36] and radar [37] sensors. The main characteristic of these models is the ability to simultaneously extract features optimized for image classification.…”

Remote Sensing for the Quantification of Land Surface Dynamics in Large River Delta Regions—A Review

“…Poursanidis et al [194] tested the suitability of the Sentinel-2 coastal aerosol band (443 nm) for the mapping of coastal underwater habitats with high spatial variability (e.g., seagrass) and shallow water bathymetry, showing promising initial results. Notably, using a convolutional neural network, Feng et al [195] fused multitemporal Sentinel-1 SAR and Sentinel-2 optical imagery to map land cover in the Yellow River Delta National Nature Reserve with an overall accuracy of 93.78%.…”

Long-Term Mapping of a Greenhouse in a Typical Protected Agricultural Region Using Landsat Imagery and the Google Earth Engine

“…We view the greenhouse as a typical object of Human-Environment interaction in agriculture [62], and our research aims to achieve a better understanding of the past, present, and future of it. Future work should consider several aspects, including (1) expanding the study area of the greenhouse to achieve a more macro view (e.g., at provincial or national scale) with more advanced classification approach like deep learning [63]; (2) seeking deep insight into market demand, government planning, human decision and their consequences on the greenhouse by connecting socio-economic data with our spatial data using Meta-analysis [64]; (3) simulating the greenhouse dynamics by coupling top-down (e.g., agent-based model [65]) and bottom-up (e.g., cellular automata [66]) strategies for different hypothetical scenarios in the future.…”