Deep Learning–Based Land Use and Land Cover Classification Using the Eurosat Dataset

Land use and land cover (LULC) classification is essential for environmental monitoring, urban planning, and resource management. This study explores the performance of three state-of-the-art deep learning architectures, MobileNetV3, ResNet34, and GoogleNet, which were enhanced with transfer learning, data augmentation, and adaptive learning rate scheduling. We evaluate these models on two benchmark datasets: EuroSAT, consisting of Sentinel-2 satellite imagery across 10 land cover classes, and PatternNet, a high-resolution aerial dataset with 38 diverse classes. The results demonstrate that MobileNetV3 achieved the highest overall accuracy (97.83% on EuroSAT and 99.23% on PatternNet) with minimal inference time, making it ideal for real-time applications. ResNet34 achieved 97.56% and 99.06% accuracy, respectively, excelling in classifying complex, visually similar classes due to its residual learning blocks. GoogleNet's balanced performance and efficiency achieved 97.36% and 99.58% accuracy across both datasets. An ablation study confirmed that data augmentation, transfer learning, and learning rate scheduling contributed to improvements in accuracy of 5-13%. This research highlights the effectiveness of modern deep learning architectures and optimized training pipelines for LULC classification across diverse datasets, providing a foundation for future advancements in cross-domain remote sensing applications.

Currently, remote sensing images (RSIs) are often exploited in the explanation of urban and rural areas, change recognition, and other domains. As the majority of RSI is high-resolution and contains wide and varied data, proper interpretation of RSIs is most important. Land use and land cover (LULC) classification utilizing deep learning (DL) is a common and efficient manner in remote sensing and geospatial study. It is very important in land planning, environmental monitoring, mapping, and land management. But, one of the recent approaches is problems like vulnerability to noise interference, low classification accuracy, and worse generalization ability. DL approaches, mostly Convolutional Neural Networks (CNNs) revealed impressive performance in image recognition tasks, making them appropriate for LULC classification in RSIs. Therefore, this study introduces a novel Land Use and Land Cover Classification employing the River Formation Dynamics Algorithm with Deep Learning (LULCC-RFDADL) technique on RSIs. The main objective of the LULCC-RFDADL methodology is to recognize the diverse types of LC on RSIs. In the presented LULCC-RFDADL technique, the dense EfficientNet approach is applied for feature extraction. Furthermore, the hyperparameter tuning of the Dense EfficientNet method was implemented using the RFDA technique. For the classification process, the LULCC-RFDADL technique uses the Multi-Scale Convolutional Autoencoder (MSCAE) model. At last, the seeker optimization algorithm (SOA) has been exploited for the parameter choice of the MSCAE system. The achieved outcomes of the LULCC-RFDADL algorithm were examined on benchmark databases. The simulation values show the better result of the LULCC-RFDADL methods with other approaches in terms of different metrics.

Efficiently implementing remote sensing image classification with high spatial resolution imagery can provide significant value in land use and land cover (LULC) classification. The new advances in remote sensing and deep learning technologies have facilitated the extraction of spatiotemporal information for LULC classification. Moreover, diverse disciplines of science, including remote sensing, have utilised tremendous improvements in image classification involving convolutional neural networks (CNNs) with transfer learning. In this study, instead of training CNNs from scratch, the transfer learning was applied to fine-tune pre-trained networks Visual Geometry Group (VGG16) and Wide Residual Networks (WRNs), by replacing the final layers with additional layers, for LULC classification using the red–green–blue version of the EuroSAT dataset. Moreover, the performance and computational time are compared and optimised with techniques such as early stopping, gradient clipping, adaptive learning rates, and data augmentation. The proposed approaches have addressed the limited-data problem, and very good accuracies were achieved. The results show that the proposed method based on WRNs outperformed the previous best results in terms of computational efficiency and accuracy, by achieving 99.17%.

The classification of land use and land cover (LULC) holds significant importance in the effective management of natural resources and the understanding of landscape dynamics in response to climatic factors. By employing high-resolution satellite imagery and a Geographical Information System (GIS), the monitoring of rapid changes in Land Use and Land Cover (LULC) becomes a straightforward task. The focus of this study is the Cameron Highlands in Malaysia, an area that experiences significant levels of land encroachment, notably in the agricultural and urban domains. The primary aim of this study is to evaluate the effectiveness of land use and land cover (LULC) classification by employing two separate satellite datasets, SPOT-7 and Sentinel-2. The classification process will be conducted using ArcGIS software. The supervised classification technique involves the utilisation of the maximum likelihood image classification algorithm to generate eight land use and land cover (LULC) categories based on the satellite images obtained in 2022. The eight categories include agricultural canopy, bare land, built-up regions, crop and farm, forest, grassland, tea plantation, and aquatic bodies. According to the result derived by SPOT-7 data classification, there was a 17.67% increase in agricultural canopy and built-up by 3.8%. In contrast, the Sentinel-2 data shows higher increases of 19.9% in agricultural canopy and 4% in built-up areas. The reduction in forest cover was the most significant consequence of the growth of agricultural canopy areas, with crop and farmland, grassland, and barren land experiencing subsequent decreases. The findings indicate that the use of Sentinel-2 images resulted in marginally superior overall accuracy and Kappa coefficient values at 90.3% and 0.853, respectively. Furthermore, SPOT-7 exhibits comprehensive accuracy rate of 86.9% and a Kappa coefficient of 0.736%. The findings of this study illustrate the effectiveness of utilising both Sentinel-2 and SPOT-7 imagery in accurately mapping land use and land cover (LULC) classifications. The inclusion of upper-level clouds and their corresponding shadows in the analysis has the potential to negatively impact the accuracy of high-resolution SPOT-7 data classification.

- Accurate classification of Land Use and Land Cover (LULC) is fundamental to understanding the spatial distribution of natural and anthropogenic features on the Earth's surface. It provides essential insights for urban planning, agricultural development, environmental monitoring, and resource management. The rapid pace of urbanization—particularly in developing regions—has amplified the demand for timely and precise LULC data. Traditional methods, such as manual interpretation and field surveys, are increasingly inadequate due to limitations in scalability, efficiency, and consistency. This study proposes an automated LULC classification approach that leverages deep learning and remote sensing technologies. Utilizing the ResNet50 deep convolutional neural network and the EuroSAT dataset comprising multispectral satellite imagery, the model is trained to classify land cover types such as urban areas, vegetation, water bodies, agricultural zones, and barren land. The classification process involves tiling satellite images into smaller segments, enabling fine-grained spatial pattern detection and high-resolution mapping. The resulting LULC maps visualize land cover categories with colorcoded tiles, facilitating rapid and accurate assessments. This approach demonstrates notable improvements in classification speed, accuracy, and consistency, making it suitable for regular environmental monitoring. By integrating artificial intelligence with satellite imagery, the proposed system offers a scalable solution for informed decision-making in land management, sustainability planning, and urban development. As remote sensing data becomes increasingly accessible and frequent, deep learning-based LULC classification systems will play a pivotal role in addressing contemporary environmental and urban challenges.

Recent advances in land use and land cover (LULC) classification with remote sensing imagery are driven by state-of-the-art models such as Convolutional Neural Networks (CNNs). Advanced CNN architecture like ResNet can enhance overall classification performance by incorporating residual skip connections. The integration of 3D feature extraction and ResNet architecture suggests a potential improvement in classification tasks. This paper explores the potential of the 3D ResNet model for LULC classification, comparing it with baseline approaches (Support Vector Machine, Random Forest, XGBoost, 1D CNN, 3D CNN) and state-of-the-art 3D models (3D VGG, 3D DenseNet) using WorldView-2 satellite imagery. The 3D ResNet-18 model, fine-tuned via transfer learning on multispectral images, demonstrates significant improvements in classification performance over machine learning models. It achieves the highest Overall Accuracy (OA) of 99.66% and Kappa Accuracy (KA) of 99.39% on the primary dataset. Despite having slightly lower performance on the external validation dataset (OA:82.89%, KA:80.05%) than 3D DenseNet, it is highly efficient with processing times of 490.2 minutes and 3.6 minutes for both datasets respectively. McNemar’s test results show 3D ResNet and 3D DenseNet have significant differences in classification performance (p<0.05) against other models consistently for both datasets.

Landsat-like moderate resolution remote sensing images are widely used in land use and land cover (LULC) classification. Limited by coarser resolutions, most of the traditional LULC classifications that are based on moderate resolution remote sensing images focus on the spectral features of a single pixel. Inspired by the spatial evaluation methods in landscape ecology, this study proposed a new method to extract neighborhood characteristics around a pixel for moderate resolution images. 3 landscape-metric-like indexes, i.e., mean index, standard deviation index, and distance weighted value index, were defined as adjacent region features to include the surrounding environmental characteristics. The effects of the adjacent region features and the different feature set configurations on improving the LULC classification were evaluated by a series of well-controlled LULC classification experiments using K nearest neighbor (KNN) and support vector machine (SVM) classifiers on a Landsat 8 Operational Land Imager (OLI) image. When the adjacent region features were added, the overall accuracies of both the classifiers were higher than when only spectral features were used. For the KNN and SVM classifiers that used only spectral features, the overall accuracies of the LULC classification were 85.45% and 88.87%, respectively, and the accuracies were improved to 94.52% and 96.97%. The classification accuracies of all the LULC types improved. Highly heterogeneous LULC types that are easily misclassified achieved greater improvements. As comparisons, the grey-level co-occurrence matrix (GLCM) and convolutional neural network (CNN) approaches were also implemented on the same dataset. The results revealed that the new method outperformed GLCM and CNN approaches and can significantly improve the classification performance that is based on moderate resolution data.

Land Use and Land Cover (LULC) are key indicators of global environmental change. As a result, the extensive effort was dedicated to creating larger-scale products of LULC from Remote Sensing (RS) data, allowing the technical group to utilize these products for a wide array of downstream applications. This phenomenon causes widespread anxiety about natural resources. Therefore, observing LULC changes was significant for natural resource management and evaluating the effects of environmental change. Machine Learning (ML) has recently gained significance for fast and accurate LULC mapping using RS data, driven by the growing requirement for ecological, environmental, and resource management. It is crucial to compute the performance of diverse ML models for reliable LULC mapping. This study proposes a novel Remote Sensing-Based Land Use and Land Cover Classification Using Deep Learning with Tuna Swarm Optimisation (RSLULCC-DLTSO) methodology. The RSLULCC-DLTSO methodology aims to advance intelligent and automated LULC classification systems that assist in sustainable land management and environmental decision-making. In the pre-processing stage, the RSLULCC-DLTSO technique utilizes a Wiener Filtering (WF) model to eliminate noise and enhance the quality of satellite images. Furthermore, the DenseNet-121-based feature extraction captures hierarchical spatial patterns and textures from RSI. A Variational Autoencoder (VAE) model is also used for LULC classification. Finally, the Tuna Swarm Optimisation (TSO) model optimally adjusts the hyperparameter values of the VAE technique, resulting in improved classification performance. A wide range of simulation analyses of the RSLULCC-DLTSO approach is implemented under the EuroSat dataset. The comparative study of the RSLULCC-DLTSO approach illustrated a superior accuracy value of 98.57% compared to existing models.

Land use and land cover (LULC) classification is becoming faster and more accurate thanks to new deep learning algorithms. Moreover, new high spectral- and spatial-resolution datasets offer opportunities to classify land cover with greater accuracy and class specificity. However, deploying deep learning algorithms to characterize present-day, modern land cover based on state-of-the-art data is insufficient for understanding trends in land cover change and identifying changes in and drivers of ecological and social variables of interest. These identifications require characterizing past land cover, for which imagery is often lower-quality. We applied a deep learning pipeline to classify land cover from historical, low-quality RGB aerial imagery, using a case study of Vancouver, Canada. We deployed an atrous convolutional neural network from DeepLabv3+ (which has previously shown to outperform other networks) and trained it on modern Maxar satellite imagery using a modern land cover classification. We fine-tuned the resultant model using a small dataset of manually annotated and augmented historical imagery. This final model accurately predicted historical land cover classification at rates similar to other studies that used high-quality imagery. These predictions indicate that Vancouver has lost vegetative cover from 1995–2021, including a decrease in conifer cover, an increase in pavement cover, and an overall decrease in tree and grass cover. Our workflow may be harnessed to understand historical land cover and identify land cover change in other regions and at other times.

Agricultural land use and land cover (LULC) classification using synthetic aperture radar (SAR) data is a fundamental application in remote sensing and precision agriculture. Leveraging the abilities of SAR, which can enter over cloud cover and deliver detailed data about surface features, allows a robust analysis of agricultural landscapes. By harnessing the control of SAR data and innovative deep learning (DL) methods, this technique provides a complete solution for effectual and automatic agricultural land classification, paving the method for informed decision-making in present farming systems. This study introduces a new gradient based optimizer with deep learning based agricultural land use and land cover classification (GBODL-ALULC) technique on SAR data. The GBODL-ALULC technique aims to detect and classify distinct types of land cover that exist in the SAR data. In the GBODL-ALULC technique, the feature extraction process takes place by a residual network with a convolutional block attention mechanism (ResNet-CBAM) model. At the same time, the GBO system has been executed for the best hyperparameter choice of the ResNet-CBAM model which helps to improve the overall LULC classification results. Finally, a regularized extreme learning machine (RELM) algorithm has been for the detection and classification of land covers. The performance study of the GBODL-ALULC method is carried out on the SAR dataset. The simulation outcome depicted that the GBODL-ALULC methodology reaches effectual LULC classification outcomes over compared methods.

Land Use and Land Cover (LULC) classification using remote sensing data is a challenging problem that has evolved with the update and launch of new satellites in orbit. As new satellites are launched with higher spatial and spectral resolution and shorter revisit times, LULC classification has evolved to take advantage of these improvements. However, these advancements also bring new challenges, such as the need for more sophisticated algorithms to process the increased volume and complexity of data. In recent years, deep learning techniques, such as convolutional neural networks (CNNs), have shown promising results in this area. Training deep learning models with complex architectures require cutting-edge hardware, which can be expensive and not accessible to everyone. In this study, a simple CNN based on the LeNet architecture is proposed to perform LULC classification over Sentinel-2 images. Simple CNNs such as LeNet require less computational resources compared to more-complex architectures. A total of 11 LULC classes were used for training and validating the model, which were then used for classifying the sub-basins. The analysis showed that the proposed CNN achieved an Overall Accuracy of 96.51% with a kappa coefficient of 0.962 in the validation data, outperforming traditional machine learning methods such as Random Forest, Support Vector Machine and Artificial Neural Networks, as well as state-of-the-art complex deep learning methods such as ResNet, DenseNet and EfficientNet. Moreover, despite being trained in over seven million images, it took five h to train, demonstrating that our simple CNN architecture is only effective but is also efficient.

Land use and land cover mapping is very important in the fields of urban planning, land management, and natural resource conservation. Recently, convolutional neural networks (CNNs) are applied widely in land use and land cover (LULC) classification as the acquisition of high resolution satellite images becomes easier owing to technological advancements. In this paper, we explore how to better exploit existed CNNs in LULC classification task. Three different learning modalities: full-trained, fine-tuning and pre-trained CNNs were used as feature extractors, and two promising CNNs models (AlexNet and GoogLeNet) and two remote sensing datasets (UC Merced Land Use dataset and Brazilian Coffee Scenes dataset) were studied. Results show that the both AlexNet and GoogLeNet can be used in high remote sensing classification with a great performance. What is more, the full-trained CNNs is not always the best approach, on the contrary, fine-tuning based on pre-trained CNNs, tends to be the best approach.

The European Commission launch of the twin Sentinel-2 satellites provides new opportunities for land use and land cover (LULC) classification because of the ready availability of their data and their enhanced spatial, temporal and spectral resolutions. The rapid development of machine learning over the past decade led to data-driven models being at the forefront of high accuracy predictions of the physical world. However, the contribution of the driving variables behind these predictions cannot be explained beyond generalized metrics of overall performance. Here, we compared the performance of three shallow learners (support vector machines, random forest, and extreme gradient boosting) as well as two deep learners (a convolutional neural network and a residual network with 50 layers) in and around the city of Malmö in southern Sweden. Our complete analysis suite involved 141 input features, 85 scenarios, and 8 LULC classes. We explored the interpretability of the five learners using Shapley additive explanations to better understand feature importance at the level of individual LULC classes. The purpose of class-level feature importance was to identify the most parsimonious combination of features that could reasonably map a particular class and enhance overall map accuracy. We showed that not only do overall accuracies increase from shallow (mean = 84.64%) to deep learners (mean = 92.63%) but that the number of explanatory variables required to obtain maximum accuracy decreases along the same gradient. Furthermore, we demonstrated that class-level importance metrics can be successfully identified using Shapley additive explanations in both shallow and deep learners, which allows for a more detailed understanding of variable importance. We show that for certain LULC classes there is a convergence of variable importance across all the algorithms, which helps explain model predictions and aid the selection of more parsimonious models. The use of class-level feature importance metrics is still new in LULC classification, and this study provides important insight into the potential of more nuanced importance metrics.

Aquatic environments and ecosystems are significantly influenced by riparian land use and land cover (LULC) patterns.With the continuous increase in human activities, LULC patterns are undergoing constant transformation, subsequently altering their impacts on aquatic environments.To proactively respond to changes in aquatic environments, continuous monitoring of riparian LULC changes is essential.Historically, LULC monitoring has been conducted primarily through field surveys, which present clear temporal and financial limitations when investigating extensive areas.Remote sensing techniques have emerged as viable alternatives to field surveys due to their capacity to provide information across broad geographical regions.When integrated with artificial intelligence technologies such as deep learning, these techniques enable area-based LULC monitoring.The accuracy of remote sensing and artificial intelligence (AI)-based LULC monitoring depends heavily on the performance of trained models, highlighting the importance of developing high-performance models.However, in real-world environments, LULC data often exhibits class imbalance, with unequal representation across categories.This data imbalance diminishes model performance, ultimately reducing the accuracy of LULC monitoring.In this study, we developed a U-Net-based riparian LULC classification algorithm for the main Nakdong River basin in South Korea, implementing loss functions that account for data imbalance.We evaluated the performance of various imbalance-addressing loss functions-Combo loss, Focal loss, Dice loss, and Tversky lossagainst the standard sparse categorical cross-entropy (SCCE) loss that does not address imbalance.Performance evaluation revealed that the U-Net model incorporating Combo loss demonstrated the highest performance (F-1 score = 0.8529, intersection over union [IoU]=0.7519),while the U-Net model with Tversky loss (F-1 score = 0.8426, IoU=0.7357) also outperformed the model using SCCE loss (F-1 score = 0.8375, IoU=0.7296).However, U-Net models employing Focal loss and Dice loss showed inferior Yan et al. (2023) Taihu Sentinel-2 convolutional neural network (CNN)

One of the materials essential for human life that must manage properly is the land. Land use and land cover (LULC) classification can help us how to manage land. The satellite can record images that can use as the data for LULC classification. This research aims to perform LULC classification using Convolutional Neural Network (CNN) on EuroSAT remote sensing image dataset taken from the Sentinel-2 satellite. CNN has become a well-known method to deal with image feature extraction. We used several CNN for feature extraction, such as VGG19, ResNet50, and InceptionV3. Then, we recalibrated the feature of CNN using Channel Squeeze & Spatial Excitation (sSE) block. We also used Support Vector Machine (SVM) and Twin SVM (TWSVM) as the classifier. VGG19 with sSE block and TWSVM achieved the highest experimental results with 94.57% accuracy, 94.40% precision, 94.40% recall, and 94.39% F1-score.