Diffusion model-based data augmentation for land cover segmentation in Pol-SAR imagery

The use of convolutional neural networks improved greatly data synthesis in the last years and have been widely used for data augmentation in scenarios where very imbalanced data is observed, such as land cover segmentation. Balancing the proportion of classes for training segmentation models can be very challenging considering that samples where all classes are reasonably represented might constitute a small portion of a training set, and techniques for augmenting this small amount of data such as rotation, scaling and translation might be not sufficient for efficient training. In this context, this paper proposes a methodology to perform data augmentation from few samples to improve the performance of CNN-based land cover semantic segmentation. First, we estimate the latent data representation of selected training samples by means of a mixture of Gaussians, using an encoder-decoder CNN. Then, we change the latent embedding used to generate the mixture parameters, at random and in training time, to generate new mixture models slightly different from the original. Finally, we compute the displacement maps between the original and the modified mixture models, and use them to elastically deform the original images, creating new realistic samples out of the original ones. Our disentangled approach allows the spatial modification of displacement maps to preserve objects where deformation is undesired, like buildings and cars, where geometry is highly discriminant. With this simple pipeline, we managed to augment samples in training time, and improve the overall performance of two basal semantic segmentation CNN architectures for land cover semantic segmentation.

This study investigated a deep learning method for fluvial land cover classification using aerial imagery of the UAV (Unmanned Aerial Vehicles). The deep learning used in this study was the CNN-Supervised Classification (CSC) developed by Carbonneau et al. (Remote Sensing of Environment, 2020). The analysis was based on 51 river sections of RGB orthorectified images taken aerially in 2015-2019 in several river channels of the Kinu River in Japan. They were obtained by applying the SfM (Structure from Motion) processing to the UAV aerial images acquired during field observations. The spatial resolution of the images was approximately 4 cm per pixel. The seven land cover types classified by CSC were water surface, gravel, sand, grass, tree, farmland, and artificial land. The deep learning algorithm CSC in this study was a classification model combining two-stage Convolution Neural Networks (CNNs). The first stage of the CSC classified the input image into 200 x 200-pixel image tiles and created a training dataset to be used in the second stage. Then, the training dataset was used to train a second-stage small-scale CNN (hereafter called mini-CNN) to optimise the model hyper-parameters. Finally, the trained CSC performed pixel-based land cover classification of the RGB orthoimages. In the first stage, this study used an existing CNN architecture, VGG16. The fine-tuning dataset had more than 2,500 images for each land cover class, resulting in a total of 85,800 through data augmentation. The hyper-parameters examined were the learning rate, patch size and the number of frozen layers. The F-measures for the CSC first stage with the optimised parameters were 99.1, 96.9, 92.6, 91.4, 93.6, 95.7 and 96.1% for water surface, gravel, sand, grass, tree, farmland, and artificial land, respectively. Then, the architecture of the mini-CNN, learning rate, patch size, patch number and filter size were optimised for the CSC second stage. The weighted average F-measure for the optimised CSC model was 90.4%. This confirmed that the optimised CSC could reproduce the land cover classes with enough accuracy. The CSC application to the RGB orthorectified images of the Kinu River in Japan showed that the CSC deep learning method could accurately classify temporal changes in fluvial geomorphologies such as gravel beds and sandbars as well as riparian vegetation, including the significant differences before and after the severe floods in 2015 and 2019. Future work would be needed to verify the applicability of the proposed CSC deep learning method to other rivers with different fluvial characteristics.

Populations in areas with limited hydrological data face ongoing challenges related to water supply and management, with climate change increasing the risks of floods and droughts. New remote sensing and modeling tools can improve land and water management in these regions, especially when combined with limited ground measurements and local knowledge of extreme events. This study examined hydrological extremes and land cover change impacts in the Grande Rivière du Nord watershed, Haiti, using satellite and model-based data. Precipitation extremes were obtained from the Global Precipitation Measurement Integrated Multi-satellite Retrievals for GPM (GPM IMERG; 2000–2025), and streamflow data were sourced from the Group on Earth Observation Global Water Sustainability (GEOGLOWS) system and bias-corrected with a small historical hydrologic database. Annual maximum series were created and fitted with Gumbel, Lognormal, and Generalized Extreme Value (GEV) distributions using the L-moment method. Goodness-of-fit tests identified the best models, and precipitation amounts for return periods of 2–100 years were estimated. The precipitation maxima aligned with locally reported extreme events, and GEV provided the best overall fit. Using the bias-corrected streamflow, a hydrologic model was calibrated and validated and then applied to land cover change scenarios. Simulations suggest that moderate land-use change can increase peak flows beyond channel capacity, raising flood risk and informing adaptation planning in northern Haiti, which has limited data.

A driver stress detection model via data augmentation based on deep convolutional recurrent neural network

National Forest Inventories (NFIs) provide valuable land cover (LC) information but often lack spatial continuity and an adequate update frequency. Satellite-based remote sensing offers a viable alternative, employing machine learning to extract thematic data. State-of-the-art methods such as convolutional neural networks rely on fully pixel-level annotated images, which are difficult to obtain. Although reference LC datasets have been widely used to derive annotations, NFIs consist of point-based data, providing only sparse annotations. Weakly supervised and self-supervised learning approaches help address this issue by reducing dependence on fully annotated images and leveraging unlabeled data. However, their potential for large-scale LC mapping needs further investigation. This study explored the use of NFI data with deep learning and weakly supervised and self-supervised methods. Using Sentinel-2 images and the Portuguese NFI, which covers other LC types beyond forest, as sparse labels, we performed weakly supervised semantic segmentation with a convolutional neural network to create an updated and spatially continuous national LC map. Additionally, we investigated the potential of self-supervised learning by pretraining a masked autoencoder on 65,000 Sentinel-2 image chips and then fine-tuning the model with NFI-derived sparse labels. The weakly supervised baseline achieved a validation accuracy of 69.60%, surpassing Random Forest (67.90%). The self-supervised model achieved 71.29%, performing on par with the baseline using half the training data. The results demonstrated that integrating both learning approaches enabled successful countrywide LC mapping with limited training data.

To improve the accuracy of wildfire risk identification in areas adjacent to power transmission corridors, this study proposes a wildfire early warning method that integrates refined land cover segmentation and multimodal feature deep learning. First, an improved bi-branch semantic segmentation network (BuildFormer++) is used to perform refined classification of high-resolution remote sensing images, extracting six types of land cover information, including forest and cultivated land. Second, a multi-dimensional feature set integrating land cover, topography, climate, and human activities is constructed and input into a multimodal wildfire point prediction network for deep feature fusion and probabilistic modeling. Experimental results show that the proposed segmentation network achieves a mean intersection–union ratio (mIoU) of 40.68% in the semantic segmentation task; the early warning model achieves an accuracy of 85.37%, an F1 score of 93.15%, and an ROC-AUC of 85.42% in risk prediction, significantly outperforming comparative methods. The “refined segmentation–feature fusion–risk prediction” framework constructed by this method can provide reliable technical support for the operation and maintenance safety and fire prevention of power transmission corridors.

Land cover provides crucial information related to biological geography, ecological climatology, and human activities. In the past, land cover mapping was performed based on visual interpretation, but it had limitations in terms of time and cost. Recently, it has become possible to create land cover maps with higher temporal resolution over wider areas using artificial intelligence-based models. The accuracy and reliability of AI model-based land cover maps increase with the amount of training data, but it is difficult to acquire large amounts of data due to the time required for label data annotation. In South Korea, the Environmental Geographic Information Service provides self-learning data consisting of aerial orthoimages and subdivision land cover classification level label data, making it possible to collect high-quality data. Therefore, this study examined the feasibility of self-learning data by building and evaluating a U-Net-based land cover classification model for waterfront areas using self-learning data. The trained model showed relatively low performance with an F-1 score of 0.61 for training data and 0.31 for test data. The model’s low performance is thought to be due to insufficient training caused by the large number of classification categories (34) and data imbalance between categories. Although the model performance using self-learning data was low, it is believed that model performance can be improved by grouping classification categories according to research purposes or resolving data imbalance through data augmentation techniques. Therefore, self-learning data is expected to be utilized in various studies using land cover.

In the scenario of limited labeled remote-sensing datasets, the model’s performance is constrained by the insufficient availability of data. Generative model-based data augmentation has emerged as a promising solution to this limitation. While existing generative models perform well in natural scene domains (e.g., faces and street scenes), their performance in remote sensing is hindered by severe data imbalance and the semantic similarity among land-cover classes. To tackle these challenges, we propose the Multi-Class Guided GAN (MCGGAN), a novel network for generating remote-sensing images from semantic labels. Our model features a dual-branch architecture with a global generator that captures the overall image structure and a multi-class generator that improves the quality and differentiation of land-cover types. To integrate these generators, we design a shared-parameter encoder for consistent feature encoding across two branches, and a spatial decoder that synthesizes outputs from the class generators, preventing overlap and confusion. Additionally, we employ perceptual loss (LVGG) to assess perceptual similarity between generated and real images, and texture matching loss (LT) to capture fine texture details. To evaluate the quality of image generation, we tested multiple models on two custom datasets (one from Chongzhou, Sichuan Province, and another from Wuzhen, Zhejiang Province, China) and a public dataset LoveDA. The results show that MCGGAN achieves improvements of 52.86 in FID, 0.0821 in SSIM, and 0.0297 in LPIPS compared to the Pix2Pix baseline. We also conducted comparative experiments to assess the semantic segmentation accuracy of the U-Net before and after incorporating the generated images. The results show that data augmentation with the generated images leads to an improvement of 4.47% in FWIoU and 3.23% in OA across the Chongzhou and Wuzhen datasets. Experiments show that MCGGAN can be effectively used as a data augmentation approach to improve the performance of downstream remote-sensing image segmentation tasks.

Machine-learning diagnostic systems are widely used to detect abnormal conditions in electrical equipment. Training robust and accurate diagnostic systems is challenging because only small databases of abnormal-condition data are available. However, the performance of the diagnostic systems depends on the quantity and quality of the data. The training database can be augmented utilizing data augmentation techniques that generate synthetic data to improve diagnostic performance. However, existing data augmentation techniques are generic methods that do not include additional information in the synthetic data. In this paper, we develop a model-based data augmentation technique integrating computer-implementable electromechanical models. Synthetic normal- and abnormal-condition data are generated with an electromechanical model and a stochastic parameter value sampling method. The model-based data augmentation is showcased to detect an abnormal condition of a distribution transformer. First, the synthetic data are compared with the measurements to verify the synthetic data. Then, ML-based diagnostic systems are created using model-based data augmentation and are compared with state-of-the-art diagnostic systems. It is shown that using the model-based data augmentation results in an improved accuracy compared to state-of-the-art diagnostic systems. This holds especially true when only a small abnormal-condition database is available.

The use of remote sensing images for land cover analysis has broad prospects. At present, the resolution of aerial remote sensing images is getting higher and higher, and the span of time and space is getting larger and larger, therefore segmenting target objects enconter great difficulties. Convolutional neural networks are widely used in many image semantic segmentation tasks, but existing models often use simple accumulation of various convolutional layers or the direct stacking of interfeature reuse of up- and downsampling, the network very heavy. To improve the accuracy of land cover segmentation, we propose a multichannel feature fusion lozenge network. The multichannel feature fusion lozenge network (MLNet) is a three-sided network composed of three branches: one branch uses different levels of feature indexes to sample to maintain the integrity of high-frequency information; one branch focuses on contextual information and strengthens the compatibility of information within and between classes; and the last branch uses feature integration to filter redundant information based on multiresolution segmentation to extract key features. Compared with FCN, UNet, PSP, and other serial single road computing models, the MLNet, which performs feature fusion after three-way parallelism structure, can significantly improve the accuracy with only small increase in complexity. Experimental results show that the average accuracy of 85.30% is obtained on the land cover data set, which is much higher than that of 82.98% of FCN, 81.87% of UNet, 77.52% of SegNet, and 83.09% of EspNet, which proves the effectiveness of the model.

Radiation information is essential to land cover classification, but general deep convolutional neural networks (DCNNs) hardly use this to advantage. Additionally, the limited amount of available remote sensing data restricts the efficiency of DCNN models though this can be overcome by data augmentation. However, normal data augmentation methods, which only involve operations such as rotation and translation, have little effect on radiation information. These methods ignore the rich information contained in the image data. In this article, the authors propose a feasible feature-based data augmentation method, which extracts spectral features that can reflect radiation information as well as geometric and texture features that can reflect image information prior to augmentation. Through feature extraction, this method indirectly enhances radiation information and increases the utilization of image information. Classification accuracies show an improvement from 80.20% to 89.20%, which further verifies the effectiveness of this method.

Sustainable development, cultural preservation, and rural quality of life depend on rural architectural design and planning. It balances modern infrastructure needs with rural ecological and social characteristics to ensure sustainable growth. This research presents a deep learning framework based on a modified and robust vision transformer optimized with a modified firefly algorithm to perform image-based land cover segmentation in rural architectural design and planning. The proposed model formulates a high-dimensional feature space from fixed-size patches, then performs positional encodings and transformer encoder layers to maintain spatial information within these patches. Then, it embeds the self-attention mechanism of these layers to capture the interdependence of global spatial regions, which is crucial for bifurcating the region of interest. The optimization of the features and selection of hyper-parameters are performed with a modified firefly algorithm. The framework is validated using the EuroSAT benchmark dataset, derived from Sentinel-2 satellite imagery, consisting of 27,000 geo-referenced samples across 10 balanced land cover classes such as annual crop, forest, herbaceous vegetation, highway, and residential areas. Each image has a spatial resolution of 64 × 64 pixels in RGB format, ensuring consistent representation of rural and urban features. The quality of the segmented regions achieved an overall accuracy of 99.5% with a kappa coefficient of 0.81, while the mean fitness value was 4.9930 × 10–8 on the publicly available rural dataset. The results are compared with state-of-the-art, pre-trained existing models like VGG-19, ResNet-50, DarkNet-19, Inception-V3, and other reported techniques like Gaussian Naïve Bayes, geosystem approach, deep CNN and fuzzy DL hybrid with chaotic PSO that have the classification accuracy of 89.96%, 91.52%, 86% and 93.3%, respectively. The stability of the proposed model is investigated through Monte Carlo simulation based on 200 independent runs and their statistical investigations. The proposed predicted system is quite helpful for rural areas’ transformation into urbanization by mitigating climate change, natural resources and geographical characteristics.

The land use land cover dataset is created using images from ArcGIS, categorized into barren land, crop land, vegetation and water bodies. The dataset is preprocessed for accuracy and diversity, and then used for land cover segmentation and classification using a highly enhanced multi-level deep fused U-Net with dual attention framework (MLDUNet-DA). The proposed architecture includes fused dilated convolution, multi-kernel-based feature recognition and Adaptive Batch Normalization. The training efficiency is improved by an adaptive weight decay factor, and dual attention enhances classification performance to strengthen the generalization of the U-Net segmentation using a Self-adaptive Hippopotamus optimization algorithm. The model is implemented on the Python platform and analyzed for accuracy, Dice score, Jaccard, IoU (Intersection Over Union), precision, recall and F1-Measure.

Graph classification is an important task widely applied to biochemistry, social networks, and other fields. Since it is a data-dependent problem, insufficient training data will deteriorate the performance of graph classification models. To address this issue, various data augmentation methods have been proposed. However, most existing methods tend to destroy the topological features, leading to a negative impact on information propagation and semantics. While the null model generates new data with the same topological features as the original graph and helps capture the latent information based on topology. Hence, combining the null model with data augmentation for graph classification may be useful in helping models learn graph representations. This paper introduces a novel null model-based augmentation technique for graph classification. Specifically, four standard and four approximate null model-based augmentation methods are developed to verify the effectiveness of the technique. Experimental results on benchmark datasets demonstrate significant performance improvements with the proposed technique. Depending on the design mechanisms of the null models, standard augmentation methods outperform the approximate ones. These findings emphasize the critical role of non-trivial features in enhancing augmentation models for different network structures, providing a new perspective on data augmentation for studying graph classification methods.

The lack of reliable and up-to-date data in developing countries is a major obstacle to sustainable development. In Morocco, where groundwater withdrawals by farmers are very intensive and informal, maps describing and monitoring the extension of irrigated areas are scarce and labor-intensive to obtain. In this paper a novel transfer learning algorithm is proposed to map irrigated areas at different stages of an agricultural cycle from Landsat 8 images. The results obtained displays satisfactory performance over traditional machine learning algorithms. On a small dataset, we initially tested three well known deep learning architectures (SegNet, DenseNet and Unet). The results obtained were not satisfactory. So, to get high performance, we rely on a transfer learning architecture combining UNet with ResNet50 backbone (trained on 2012 ILSVRC ImageNet dataset) as a baseline after a phase where different configurations were tested. In the first part of this study, we compared the use of three optimization methods: Adam and two variants of Stochastic Gradient Descent (SGD) associated with two techniques (Cyclical Learning Rate and Warm Restart) to find the optimal learning rate and then test the impact of data augmentation on the overall accuracies. Data augmentation had improved the overall accuracy for the three methods. Adam based method from 94% to 97% with mean IoU of 0,79 (for all land cover classes) and 0,86 for irrigated areas class. For SGD based methods, the overall accuracy had increased from 91% to 94% with mean IoU of 0,75 (for all land cover classes) and 0,82 for irrigated areas class. As we are interested in having irrigated areas maps at different key periods of the agricultural cycle, we also explored, in the second part of this study, the temporal generalization of the best model.