Deep learning for urban land use category classification: A review and experimental assessment

Remote Sensing of Environment, Journal Year: 2024, Volume and Issue: 311, P. 114290 - 114290

Published: July 14, 2024

Mapping the distribution, pattern, and composition of urban land use categories plays a valuable role in understanding environmental dynamics facilitating sustainable development. Decades effort mapping have accumulated series approaches products. New trends characterized by open big data advanced artificial intelligence, especially deep learning, offer unprecedented opportunities for patterns from regional to global scales. Combined with large amounts geospatial data, learning has potential promote higher levels scale, accuracy, efficiency, automation. Here, we comprehensively review advances based research practices aspects sources, classification units, approaches. More specifically, delving into different settings on learning-based mapping, design eight experiments Shenzhen, China investigate their impacts performance terms sample, model. For each investigated setting, provide quantitative evaluations discussed inform more convincing comparisons. Based historical retrospection experimental evaluation, identify prevailing limitations challenges suggest prospective directions that could further facilitate exploitation techniques using remote sensing other spatial across various

Language: Английский

---

A Review of CNN Applications in Smart Agriculture Using Multimodal Data

Sensors, Journal Year: 2025, Volume and Issue: 25(2), P. 472 - 472

Published: Jan. 15, 2025

This review explores the applications of Convolutional Neural Networks (CNNs) in smart agriculture, highlighting recent advancements across various including weed detection, disease crop classification, water management, and yield prediction. Based on a comprehensive analysis more than 115 studies, coupled with bibliometric study broader literature, this paper contextualizes use CNNs within Agriculture 5.0, where technological integration optimizes agricultural efficiency. Key approaches analyzed involve image segmentation, regression, object detection methods that diverse data types ranging from RGB multispectral images to radar thermal data. By processing UAV satellite CNNs, real-time large-scale monitoring can be achieved, supporting advanced farm management. A comparative shows how perform respect other techniques traditional machine learning deep models processing, particularly when applied high-dimensional or temporal Future directions point toward integrating IoT cloud platforms for leveraging large language regulatory insights. Potential research emphasize improving increased accessibility hybrid modeling meet demands climate variability food security, positioning as pivotal tools sustainable practices. related repository contains reviewed articles along their publication links is made available.

Language: Английский

---

Efficient knowledge distillation for hybrid models: A vision transformer‐convolutional neural network to convolutional neural network approach for classifying remote sensing images

IET Cyber-Systems and Robotics, Journal Year: 2024, Volume and Issue: 6(3)

Published: July 10, 2024

Abstract In various fields, knowledge distillation (KD) techniques that combine vision transformers (ViTs) and convolutional neural networks (CNNs) as a hybrid teacher have shown remarkable results in classification. However, the realm of remote sensing images (RSIs), existing KD research studies are not only scarce but also lack competitiveness. This issue significantly impedes deployment notable advantages ViTs CNNs. To tackle this, authors introduce novel hybrid‐model approach named HMKD‐Net, which comprises CNN‐ViT ensemble CNN student. Contrary to popular opinion, posit sparsity RSI data distribution limits effectiveness efficiency transfer. As solution, simple yet innovative method handle variances during phase is suggested, leading substantial enhancements The assessed performance HMKD‐Net on three datasets. findings indicate outperforms other cutting‐edge methods while maintaining smaller size. Specifically, exceeds KD‐based with maximum accuracy improvement 22.8% across ablation experiments indicated, has cut down time expenses by about 80% process. study validates technique can be more effective efficient if RSIs well handled.

Language: Английский

---

A novel framework to assess apple leaf nitrogen content: Fusion of hyperspectral reflectance and phenology information through deep learning

Computers and Electronics in Agriculture, Journal Year: 2024, Volume and Issue: 219, P. 108816 - 108816

Published: March 15, 2024

Language: Английский

---

A deep learning-based super-resolution method for building height estimation at 2.5 m spatial resolution in the Northern Hemisphere

Remote Sensing of Environment, Journal Year: 2024, Volume and Issue: 310, P. 114241 - 114241

Published: June 4, 2024

Language: Английский

---

Analyzing winter-wheat biochemical traits using hyperspectral remote sensing and deep learning

Computers and Electronics in Agriculture, Journal Year: 2024, Volume and Issue: 222, P. 109026 - 109026

Published: May 15, 2024

Language: Английский

---

Global deep learning model for delineation of optically shallow and optically deep water in Sentinel-2 imagery

Remote Sensing of Environment, Journal Year: 2024, Volume and Issue: 311, P. 114302 - 114302

Published: July 4, 2024

In aquatic remote sensing, algorithms commonly used to map environmental variables rely on assumptions regarding the optical environment. Specifically, some assume that water is optically deep, i.e., influence of bottom reflectance measured signal negligible. Other opposite and are based an estimation bottom-reflected part signal. These may suffer from reduced performance when relevant not met. To address this, we introduce a general-purpose tool automates delineation deep shallow waters in Sentinel-2 imagery. This allows application for satellite-derived bathymetry, habitat identification, water-quality mapping be limited environments which they intended, thus enhance accuracy derived products. We sampled 440 images wide range coastal locations, covering all continents latitudes, manually annotated 1000 points each image as either or by visual interpretation. dataset was train six machine learning classification models - Maximum Likelihood, Random Forest, ExtraTrees, AdaBoost, XGBoost, neural networks utilizing both original top-of-atmosphere atmospherically corrected datasets. The were trained features including kernel means standard deviations band, well geographical location. A network emerged best model, with average 82.3% across two datasets fast processing time. Higher accuracies can achieved removing pixels intermediate probability scores predictions. made this model publicly available Python package. represents substantial step toward automatic imagery, sensing community downstream users ensure algorithms, such those bathymetry quality, applied only intended.

Language: Английский

---

Leveraging Remote Sensing Data for Yield Prediction with Deep Transfer Learning

Sensors, Journal Year: 2024, Volume and Issue: 24(3), P. 770 - 770

Published: Jan. 24, 2024

Remote sensing data represent one of the most important sources for automized yield prediction. High temporal and spatial resolution, historical record availability, reliability, low cost are key factors in predicting yields around world. Yield prediction as a machine learning task is challenging, reliable ground truth difficult to obtain, especially since new points can only be acquired once year during harvest. Factors that influence annual plentiful, acquisition expensive, crop-related often need captured by experts or specialized sensors. A solution both problems provided deep transfer based on remote data. Satellite images free charge, allows recognition yield-related patterns within countries where plentiful transfers knowledge other domains, thus limiting number observations needed. Within this study, we examine use prediction, preprocessing towards histograms unique. We present framework demonstrate its successful application gained from US soybean Argentina. perform alignment two domains improve applying several techniques, such L2-SP, BSS, layer freezing, overcome catastrophic forgetting negative problems. Lastly, exploit spatio-temporal Gaussian process. able performance Argentina total 19% terms RMSE 39% R2 compared predictions without processes. This proof concept advanced techniques form enable emerging developing countries, usually limited.

Language: Английский

---

High-Accuracy Mapping of Soil Organic Carbon by Mining Sentinel-1/2 Radar and Optical Time-Series Data with Super Ensemble Model

Remote Sensing, Journal Year: 2025, Volume and Issue: 17(4), P. 678 - 678

Published: Feb. 17, 2025

Accurate digital soil organic carbon mapping is of great significance for regulating the global cycle and addressing climate change. With advent remote sensing big data era, multi-source multi-temporal techniques have been extensively applied in Earth observation. However, how to fully mine time-series high-accuracy SOC remains a key challenge. To address this challenge, study introduced new idea mining data. We used 413 topsoil samples from southern Xinjiang, China, as an example. By (Sentinel-1/2) 2017 2023, we revealed temporal variation pattern correlation between Sentinel-1/2 SOC, thereby identifying optimal time window monitoring using integrating environmental covariates super ensemble model, achieved Southern China. The results showed following aspects: (1) windows were July–September July–August, respectively; (2) modeling accuracy sensor integrated with was superior single-source alone. In model based on data, cumulative contribution rate Sentinel-2 51.71% higher than that Sentinel-1 data; (3) stacking model’s predictive performance outperformed weight average simple models. Therefore, covariates, driven represents strategy mapping.

Language: Английский

---

A global Swin-Unet Sentinel-2 surface reflectance-based cloud and cloud shadow detection algorithm for the NASA Harmonized Landsat Sentinel-2 (HLS) dataset

Science of Remote Sensing, Journal Year: 2025, Volume and Issue: unknown, P. 100213 - 100213

Published: Feb. 1, 2025

Language: Английский

---