A Comprehensive Review of Deep Learning: Architectures, Recent Advances, and Applications

Information, Journal Year: 2024, Volume and Issue: 15(12), P. 755 - 755

Published: Nov. 27, 2024

Deep learning (DL) has become a core component of modern artificial intelligence (AI), driving significant advancements across diverse fields by facilitating the analysis complex systems, from protein folding in biology to molecular discovery chemistry and particle interactions physics. However, field deep is constantly evolving, with recent innovations both architectures applications. Therefore, this paper provides comprehensive review DL advances, covering evolution applications foundational models like convolutional neural networks (CNNs) Recurrent Neural Networks (RNNs), as well such transformers, generative adversarial (GANs), capsule networks, graph (GNNs). Additionally, discusses novel training techniques, including self-supervised learning, federated reinforcement which further enhance capabilities models. By synthesizing developments identifying current challenges, insights into state art future directions research, offering valuable guidance for researchers industry experts.

Language: Английский

---

Land Cover Classification Model Using Multispectral Satellite Images Based on a Deep Learning Synergistic Semantic Segmentation Network

Sensors, Journal Year: 2025, Volume and Issue: 25(7), P. 1988 - 1988

Published: March 22, 2025

Land cover classification (LCC) using satellite images is one of the rapidly expanding fields in mapping, highlighting need for updating existing computational methods. Advances technology and increasing variety applications have introduced challenges, such as more complex classes a demand greater detail. In recent years, deep learning Convolutional Neural Networks (CNNs) significantly enhanced segmentation images. Since training CNNs requires sophisticated expensive hardware significant time, pre-trained networks has become widespread image. This study proposes hybrid synergistic semantic method based on Deeplab v3+ network clustering-based post-processing scheme. The proposed accurately classifies various land (LC) types multispectral images, including Pastures, Other Built-Up Areas, Water Bodies, Urban Grasslands, Forest, Farmland, Others. scheme includes spectral bag-of-words model K-medoids clustering to refine outputs correct possible errors. simulation results indicate that combining with improves Matthews correlation coefficient (MCC) by approximately 5.7% compared baseline method. Additionally, approach robust data imbalance cases can dynamically update its codewords over different seasons. Finally, was several state-of-the-art methods Italy's Lake Garda (Lago di Garda) region. showed outperformed best techniques at least 6% terms MCC.

Language: Английский

---

Study on soil salinity inversion of different crop types based on multi-time series

Plant and Soil, Journal Year: 2025, Volume and Issue: unknown

Published: April 16, 2025

Language: Английский

---

Neural Networks for Analyzing Soil Organic Carbon Storage

IGI Global eBooks, Journal Year: 2025, Volume and Issue: unknown, P. 455 - 480

Published: April 11, 2025

Soil organic carbon (SOC) is an essential element of the global cycle, serving a central role in climate change mitigation, soil fertility, and ecosystem sustainability. Conventional SOC estimation techniques are time-consuming, labor-intensive, geographically confined, thus confining their efficiency for large-scale monitoring. This chapter discusses how artificial neural networks, such as CNNs, RNNs, deep learning models, improve forecasting accuracy scalability. With integration remote sensing, geospatial data, environmental factors, AI-based models facilitate effective processing mapping distribution. Deep machine methodologies enhance predictive power, automate analysis, mitigate uncertainties estimation. Critical methodologies, issues, emerging trends exploiting networks storage discussed, prioritizing sequestration monitoring optimization, sustainable land management, resilience planning.

Language: Английский

---

Soil Organic Carbon Monitoring and Modelling via Machine Learning Methods Using Soil and Remote Sensing Data

Agriculture, Journal Year: 2025, Volume and Issue: 15(9), P. 910 - 910

Published: April 22, 2025

(1) Background: Soil organic carbon (SOC) is an important parameter of soils and a critical factor in global cycling. The accurate monitoring modelling SOC are essential for assessing soil fertility, facilitating sustainable land management, mitigating climate change. (2) Methods: This research paper explores the integration machine learning (ML) approaches with soil, terrain remotely sensed data to enhance estimation. Various ML models, including Neural Networks (NNs), Random Forests (RFs), Support Vector Machines (SVMs) Decision Trees (DTs), were trained evaluated using dataset comprising laboratory data, Sentinel-2 spectral indices, topographic features. Feature selection techniques applied indicate most predictors, improving model performance interpretability. (3) Results: results demonstrate potential ML-driven achieve high accuracy prediction. (4) Conclusions: highlights advantages leveraging big artificial intelligence monitoring, providing scalable cost-effective framework assessment agricultural environmental applications.

Language: Английский

---

Remote Sensing-Based Soil Organic Carbon Monitoring Using Advanced Machine Learning Techniques Under Conservation Agriculture Systems

Smart Agricultural Technology, Journal Year: 2025, Volume and Issue: unknown, P. 101036 - 101036

Published: May 1, 2025

Language: Английский

---

Integration of remote sensing and artificial neural networks for prediction of soil organic carbon in arid zones

Frontiers in Environmental Science, Journal Year: 2024, Volume and Issue: 12

Published: Sept. 23, 2024

Introduction Mapping soil organic carbon (SOC) with high precision is useful for controlling fertility and comprehending the global cycle. Low-relief locations are characterized by minimal variability in traditional soil-forming elements, such as terrain climatic conditions, which make it difficult to reflect spatial variation of properties. In meantime, vegetation cover makes more obtain direct knowledge about agricultural soil. Crop growth biomass reflected normalized difference index (NDVI), a significant indicator. Rather than using conventional variables. Methods this study, novel model predicting SOC was developed Landsat-8 Operational Land Imager (OLI) band data (Blue (B), Green (G), Red (R), Near Infrared (NIR), NDVI supporting variables, Artificial Neural Networks (ANNs). A total 120 surface samples were collected at depth 25 cm northeastern Nile Delta near Damietta City. Of these, 80% (96 samples) randomly selected training, while remaining 24 used testing validation. Additionally, Gaussian Process Regression (GPR) models trained estimate levels Matern 5/2 kernel within Learner framework. Results discussion The results demonstrate that both ANN multilayer feedforward network GPR offer effective frameworks prediction. achieved an R 2 value 0.84, higher 0.89. These findings, supported visual statistical evaluations through cross-validation, confirm reliability accuracy models. Conclusion systematic application framework provides robust tool prediction, contributing sustainable management practices.

Language: Английский

---

A new approach in soil organic carbon estimation using machine learning algorithms: a study in a tropical forest in Vietnam

Journal of Forest Research, Journal Year: 2024, Volume and Issue: unknown, P. 1 - 10

Published: Dec. 2, 2024

Soil organic carbon (SOC) dataset augmentation, which enables the comprehensive monitoring of sinks at regional and global scales, is vital for cycle management soil fertility. SOC maps built by conventional laboratory or field measurements are time- cost-consuming especially difficult in forests. A new approach to build with good accuracy time efficiency promptly respond changes dynamics is, therefore, being identified. This study aimed evaluate ability estimation using a multiple linear regression model (MLR) four machine learning algorithms: artificial neural networks (ANN), support vector (SVM), random forest (RF), extreme gradient boosting (XGBoost) satellite data sources nutrient indicator find optimal method. The results indicate that SVM XGBoost models demonstrated best predictive abilities (R2 = 0.70 0.74, %RMSE 8.8 8.3, MAE 0.176 0.155) when remote sensing variables property variables, respectively. Band7_IDM, Band 5, Band4_IDM, RVI, NSMI, NDVI, 6, 7, 4 were most valuable model, while DVI, GVMI, Band4_Dive model. may be applied Vietnamese forests instead methods high 0.74).

Language: Английский

---