A Hybrid CNN-RNN Deep Learning Network for Deriving Cyclonic Change Map from Bi-Temporal SAR Images

Jenifer, A.; Natarajan, S. P.

doi:10.1007/978-981-16-6407-6_30

Cited by 2 publications

(1 citation statement)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…CNNs have the drawback of overfitting the input data, which results in inadequate performance on new datasets (Barzegar et al, 2021). Meanwhile, RNNs are applied to analyze temporal dependencies (Jenifer et al, 2022). Gradient loss is a common problem in RNNs (with gradient-based and back-propagation learning methods).…”

Section: Comparison and Analysis Of Modelsmentioning

confidence: 99%

Spatial prediction of physical and chemical properties of soil using optical satellite imagery: a state-of-the-art hybridization of deep learning algorithm

Hosseini,

Razavi-Termeh,

Sadeghi-Niaraki

et al. 2023

Front. Environ. Sci.

View full text Add to dashboard Cite

This research aimed to predict soil’s physical and chemical properties with a state-of-the-art hybrid model based on deep learning algorithms and optical satellite images in a region in the north of Iran. As dependent data, 317 soil samples (0–30 cm) were collected in field surveying and analyzed by the soil and water research institute for their physical (clay, silt, and sand) and chemical [electrical conductivity (EC), organic carbon (OC), phosphorus (P), soil reaction (pH), and potassium (K)] properties. Based on independent data, 23 remote sensing (RS) parameters (extracted from Landsat 8 optical images), 17 topographical parameters [extracted from the digital elevation model (DEM)], and four climatic parameters (derived from the meteorological organization). Spatial prediction of physical and chemical properties was implemented using a convolutional neural network (CNN), recurrent neural network (RNN), and hybrid CNN-RNN models. The evaluation results indicated that the hybrid CNN-RNN model had higher accuracy in all soil properties, followed by the RNN and CNN models. In the hybrid CNN-RNN model, pH (0.0206), EC (0.0958 dS/m), silt (0.0996%), P (0.1078 ppm), K (0.1185 ppm), sand (0.1360%), OC (0.1361%), and clay (0.1419%) had higher prediction accuracy, as determined by the root mean-squared error (RMSE) index. The hybrid CNN-RNN model proved to be the most effective for soil property prediction in this region. This finding underscores the potential of deep learning techniques in harnessing RS data for precise soil property mapping, with implications for land management and agricultural practices.

show abstract

Section: Comparison and Analysis Of Modelsmentioning

confidence: 99%