Search Results (416)

Background and Objective: When hypertensive cerebral hemorrhage causes secondary intraventricular hemorrhage, it usually significantly increases the complexity of the patient’s condition and the risk of poor prognosis. Deep learning methods can automatically and quickly segment intraparenchymal hemorrhage and intraventricular hemorrhage, and quantitatively analyze hematoma-related properties to provide auxiliary information for subsequent diagnosis and treatment. Methods: We retrospectively enrolled 351 patients with hypertensive cerebral hemorrhage, 219 of whom had secondary intraventricular hemorrhage. All patients underwent computed tomography within 1 week after diagnosis. Based on 3D U-Net, we developed a deep learning network with a multi-scale deformable convolution module and a softened anatomical consistency loss. The multi-scale deformable convolution module can enhance the learning ability of multi-deformation features and increase the receptive field of the network. The anatomical consistency loss, built upon softened labels, can alleviate the impact of label noise. Results: We evaluated our model at pixel, volume, and morphology levels. It achieved Dice of 0.8990 ± 0.1169 for intraparenchymal hemorrhage and 0.7124 ± 0.1227 for intraventricular hemorrhage, both higher than that of the comparison model. Compared with the Coniglobus method, our model has a narrower consistency limit and more concentrated predicted values. Additionally, in segmenting irregular and different-sized hematomas, it generates the smallest centroid, volume, position, and morphology deviations. Conclusions: The proposed model can automatically and accurately segment two types of hematomas and quantify multiple attributes, is robust in multi-deformation and label noise scenarios, and has the potential to assist in clinical diagnosis and treatment; its actual impact on clinical decision-making and patient outcomes requires prospective validation. Full article

Interpreting multispectral remote sensing data for hydrothermal alteration mapping remains challenging in complex mountainous metallogenic belts because dense pixel-level field labels are difficult to obtain and weak spectral responses are affected by lithological background, vegetation, snow/ice cover, and topographic shadow. This study proposes a principal component analysis (PCA)-guided weakly supervised workflow for mapping hydroxyl- and iron-oxide-related spectral anomalies in the Bulong–Maidan–Tuoyun gold–copper metallogenic belt, southwestern Tianshan, China, using Landsat 8 Operational Land Imager (OLI) imagery. PCA was used as a spectral prior to generate PCA-derived positive spectral anomaly samples for model training. A Residual-ECA Alteration Information Extraction (REA-AIE) model was developed to refine PCA-derived anomalies by learning local spectral–spatial features from multispectral image patches. Under the PCA-constrained random sample-level evaluation, REA-AIE achieved F1 scores of 95.90% and 97.09% for hydroxyl- and iron-oxide-related spectral anomalies, respectively; these values indicate agreement with PCA-derived pseudo-labels rather than spatially independent estimates of mapping performance. Petrography-constrained site-level assessment showed that REA-AIE-predicted spectral anomalies occurred within 90 m of 43 of the 53 altered sites, corresponding to a site-level recall of 81.13% and supporting their consistency with field-based geological evidence. Full article

Floor plan design normally depends on architectural training or CAD software, a barrier for homeowners, students, and small practices alike. The author asks a narrower question: can a general-purpose, pre-trained text-to-image model draw a usable floor plan straight from a written brief, and can a building code be folded into that process? To find out, four models, Gemini, DALL-E, DeepAI, and Stable Diffusion, were put through tests using ten descriptions of villas and apartment buildings. Relevant Saudi Building Code (SBC) clauses, covering minimum room sizes, corridor widths, accessibility, and fire-safety provisions, were retrieved and written into each prompt before generation, and the outputs were assessed quantitatively by accuracy and latency, with SBC compliance and realism recorded only as qualitative observations. Gemini was the fastest by a wide margin, averaging 8.3 s per plan against 40.5 for Stable Diffusion, the slowest, and it also scored highest for accuracy; that ordering is not established here, however, because the models were not scored by a common judge, and each description was generated only once. DALL-E drew the most realistic images but was slower and looser on detail. One limitation cut across all four: none reported room dimensions reliably, so compliance can be verified only in part from the image. A case is presented in which Gemini printed area labels directly on the image, yet a pixel-level measurement shows the room with the smallest printed area drawn as the largest of the three, an internal inconsistency that needs no external ground truth to demonstrate and that exposes the core limitation of current text-to-image decoders. On the strength of these results, the best model, Gemini, was built into a Django web application that turns a typed description into a viewable plan. The system is offered as an early-stage drafting assistant, not a code-verified architectural design tool. The study sets out plainly what off-the-shelf models and prompt-level guidance deliver without fine-tuning, and where they still fall short of professional, code-verified design. Full article

Reliable crop-type mapping from satellite image time series is affected by distribution shifts across geographic regions, agricultural years, and heterogeneous label systems. To address this challenge, we propose Cross-Domain Mixup (CDMix), a supervised domain-adaptation method designed to leverage a larger labeled source dataset to improve performance on a smaller labeled target dataset under distribution shifts. We also introduce PixelSet-Slovakia, a new multi-year, parcel-level Sentinel-2 dataset covering three Slovak study regions and several growing seasons. Using a common backbone, we compare CDMix against three families of adaptation strategies: (i) no adaptation, (ii) weight transfer through fine-tuning and encoder freezing, and (iii) feature-space alignment using Maximum Mean Discrepancy (MMD) and Correlation Alignment (CORAL). All methods are evaluated in two scenarios: geographic supervised adaptation across datasets from two countries and temporal supervised adaptation across different growing seasons. Across both tested source–target settings, CDMix generally achieves competitive performance when initialized from pretrained representations. Under the region-held-out validation protocol, several pretrained adaptation strategies outperform training from scratch. Full article

Pixel-level annotation of remote sensing imagery is costly, motivating weakly supervised semantic segmentation (WSSS) using only image-level labels. However, class activation maps (CAMs) often highlight only discriminative sub-regions and fail to separate adjacent land-cover regions, particularly in remote sensing scenes characterized by densely co-occurring land-cover classes and substantial variations in object scale. To address these limitations, we propose a three-stage framework that integrates complementary priors from Contrastive Language–Image Pre-training (CLIP), Self-Distillation with No Labels version 2 (DINOv2), and the Segment Anything Model (SAM). First, a lightweight CLIP adapter aligns vision–language priors with remote sensing imagery, while sigmoid-based multi-label decoupled distillation replaces class-competitive distillation with independent class-wise supervision, producing more complete CAMs. Second, DINOv2-guided feature clustering decomposes large merged regions before SAM prompt generation, while Spatial–Semantic Constraints are used to construct confidence-guided point-and-box prompts and reject excessively expanded or semantically inconsistent masks, thereby generating reliable pseudo-labels. Finally, a compact segmentation network is initialized with the weights learned in Stage 1 and retrained using the refined pseudo-labels generated in Stage 2, eliminating the need for foundation models during inference. Experiments on the Potsdam, LoveDA, and DeepGlobe datasets show that the proposed method achieves mean intersection over union (mIoU) scores of 53.16%, 52.66%, and 62.98%, respectively, outperforming state-of-the-art WSSS baselines by 6.55, 1.16, and 1.27 percentage points, respectively. These results demonstrate the effectiveness and generalizability of the proposed framework across diverse remote sensing scenarios under image-level supervision. Full article

Region-level weakly supervised hyperspectral target detection (HTD) using multiple-instance learning (MIL) reduces annotation costs but encounters challenges such as bag label ambiguity, boundary over-smoothing, and test-time computational latency. To address these issues, we propose Hyper-VMIL, a spatial–spectral topology-regularized variational hypergraph network. Hyper-VMIL formulates latent target localization as variational inference over dual-path hypergraphs: a boundary-aware spatial hypergraph modeling geometric patch continuity and a dynamic spectral-manifold hypergraph capturing non-local material similarity. Node-adaptive gating dynamically balances spatial and spectral evidence to mitigate over-smoothing near target boundaries. Furthermore, a confidence-aware continuous posterior refinement (CTPR) mechanism reduces the confirmation bias associated with conventional hard pseudo-label binarization. Finally, a teacher–student distillation strategy transfers contextual topology into a lightweight single-spectrum student detector. Benchmark experiments on simulated ASTER and airborne MUUFL Gulfport and Avon datasets show that Hyper-VMIL achieves competitive performance against 15 baseline methods. Notably, Hyper-VMIL supports dual inference modes: Context Mode provides improved detection accuracy (+4.6% average NAUC over VMIL-ECM on MUUFL), while Pixel Mode enables single-spectrum inference ($1.25\mathsf{\mu }\mathrm{s}$ single-instance latency and an amortized streaming throughput of $0.015\mathsf{\mu }\mathrm{s}$ per pixel) suitable for onboard real-time deployment. Full article

Mangroves are important coastal blue-carbon ecosystems, and accurate biomass estimation is essential for carbon stock assessment and ecological monitoring. To address the limitation of regional-scale biomass estimation caused by the scale mismatch between field plots and satellite pixels, this study selected the Zhangjiangkou National Mangrove Nature Reserve in Fujian Province as the study area and the mangrove distribution region of Fujian Province as the extrapolation area. A multiscale biomass estimation framework integrating field plots, unmanned aerial vehicles (UAVs), and satellite remote sensing was established. The results showed that (1) the optimal UAV-scale models achieved R² values of 0.69 and 0.78 for aboveground biomass (AGB) and belowground biomass (BGB), respectively, with corresponding root mean square error (RMSE) values of 18.55 and 9.52 t·ha⁻¹ and normalized root mean square error (nRMSE) values of 0.14 and 0.17 demonstrating reliable predictive performance; (2) after introducing UAV-derived bridging labels, the R² of the AGB model increased from 0.24 to 0.64, while the RMSE decreased from 29.02 to 10.86 t·ha⁻¹. Similarly, the R² of the BGB model increased from 0.43 to 0.63, accompanied by a reduction in RMSE from 14.89 to 6.35 t·ha⁻¹, demonstrating a substantial improvement in satellite-scale biomass estimation accuracy; (3) the total AGB and BGB of mangroves in Fujian Province were estimated at 58,768.60 t and 24,575.14 t, respectively, with high-biomass areas mainly distributed along the coastal regions of Zhangzhou and Quanzhou. Unlike conventional field-to-satellite extrapolation approaches, the proposed framework introduces UAV-derived biomass maps as intermediate bridging labels for pixel-level supervised learning, thereby establishing an effective link between field measurements and satellite observations. This strategy effectively reduces the scale mismatch between field and satellite data, significantly improves satellite-scale biomass estimation accuracy, and provides a transferable and scalable framework for regional mangrove biomass mapping and blue-carbon assessment. Full article

Remote sensing change detection technology is widely used in land-use monitoring, urban planning, and disaster assessment. However, during imaging and transmission, bi-temporal remote sensing images are vulnerable to Gaussian noise, which makes it difficult for change detection algorithms to distinguish truly changed areas from noise-affected regions. To address this issue, this study proposes an unsupervised Gaussian-noise-robust change detection algorithm, termed FRIH-SEEDSAM. The proposed method first applies the Fast and Robust Fuzzy C-Means (FRFCM) algorithm to perform noise-resistant fuzzy clustering on bi-temporal remote sensing images. To establish reliable correspondences between the clustering results, the Integrated Region Matching (IRM) algorithm is introduced to construct weighted matching relationships while reducing the influence of abnormal memberships. The change intensity of spatially corresponding pixels is then calculated to generate a more stable change intensity map. Subsequently, the change intensity map is input into the Hybrid Conditional Random Field (HCRF) to infer pixel-level change labels, where the object potential function is constructed from the segmentation results of the Superpixels Extracted via Energy-Driven Sampling (SEEDS)-guided Segment Anything Model (SEEDSAM), which uses the centroids of the SEEDS superpixel regions as point prompts for the SAM, thereby enhancing change-label consistency within the same changed object region. The experimental results show that the FRIH-SEEDSAM algorithm maintains stable change detection performance across different datasets and under varying Gaussian noise levels. It outperforms the comparison algorithms in terms of several accuracy evaluation indicators, including Kappa and F1. Furthermore, even when the Gaussian noise variance increases to 0.05, Kappa remains at 0.8 or above on multiple dataset images. Full article

Pests and diseases are key biological stress factors affecting crop yield and quality. Semantic segmentation enables pixel-level localization and severity characterization, but its performance and generalization are constrained by the high cost of high-quality pixel-level annotations, limited labeled samples, and class imbalance in agricultural datasets. We propose AgriUFM, an unconditional flow-matching framework for joint image–mask generation in agricultural pest and disease scenarios. By learning a unified continuous probability flow over the joint distribution, the framework is designed to promote structural co-evolution and spatial consistency between generated RGB images and masks. Across four evaluated datasets, AgriUFM achieved lower FID and rFID than the evaluated GAN- and diffusion-based comparators, whereas IS performance was dataset-dependent. Within the evaluated ablation configurations, uniform time sampling with 25 sampling steps and the midpoint ODE solver yielded the most favourable observed quality–efficiency trade-off. Under the held-out test protocol, the joint UFM strategy achieved higher image–mask correspondence than the M2I and I2M conditional variants. In the evaluated downstream settings, AgriUFM-generated augmentation improved MIoU and PA for U-Net and TransUNet. These results indicate that joint distribution modelling is a promising approach for structurally coherent generative augmentation in the agricultural imaging tasks studied. Full article

Riverine plastic waste is difficult to detect and map accurately because debris ranges from small items to large floating clusters, and tropical rivers present challenging conditions, such as murky water, floating vegetation, and variable lighting. This study develops and benchmarks a deep learning method for pixel-level, multi-scale mapping of riverine waste from five-band UAV multispectral imagery. We deployed a drone equipped with a five-band multispectral sensor over the Brantas River in Surabaya, East Java, Indonesia, and introduce the Multi-Scale Hierarchical Attention Ensemble (MHAE), which combines three backbone networks across three image resolutions through learned scale- and backbone-attention weighting. MHAE was evaluated against 12 CNN-, transformer-, state-space-, and traditional-machine-learning-based baselines (including Random Forest, U-Net, UNet++, and DeepLabV3+) on the accompanying BrantasRiverWaste-UAV dataset (882 image tiles from a single-site, single-season orthomosaic covering approximately 0.54 km² of river surface, labelled as water, land, organic waste, or inorganic waste), with pairwise comparisons assessed using Wilcoxon signed-rank tests with Bonferroni correction. MHAE achieved the highest pixel-level waste detection rate among the 12 evaluated models (86.76%), with a mean intersection-over-union of 78.33% (third-highest, behind UNet++ and U-Net). This work provides an initial, single-site benchmark and reference architecture for near-real-time riverine waste monitoring, and introduces the BrantasRiverWaste-UAV as, to our knowledge, one of the first five-band multispectral UAV datasets for tropical riverine waste mapping. Balanced sampling and multi-scale attention fusion can substantially improve waste-pixel detection under severe class imbalance; because the benchmark derives from a single site and season, future work should extend evaluation across additional seasons and river systems before the approach is generalised operationally. Full article

Camellia oleifera leaf disease segmentation under natural field conditions is important for precision plant protection but remains challenging because lesions often show small target areas, blurred boundaries, uneven illumination, complex backgrounds, and coexisting symptoms. To address these problems, this paper proposes BFMambaNet, a Boundary-Frequency-guided Global Semantic Mamba Network for fine-grained disease segmentation. The model adopts an encoder-decoder framework and introduces a Global Semantic Mamba-based spatial selective feature modeling block to capture long-range lesion context and reduce semantic confusion. A gated wavelet spatial enhancement block is further designed to strengthen high-frequency boundary details while suppressing noisy responses. During training, boundary-frequency auxiliary supervision guides contour localization and pathological texture recovery without additional manual boundary labels. A reinforcement-learning-guided adaptive loss controller adjusts class-wise reweighting factors and loss-component weights according to the training state, improving optimization stability. A pixel-level dataset containing 1400 images and seven disease categories was constructed for evaluation. Experimental results show that BFMambaNet achieves 92.39% Precision, 91.43% Recall, 91.26% Dice, and 85.46% mIoU, outperforming representative CNN-based, Transformer-based, and Mamba-based models. Evaluations on environmental subsets confirm superior robustness, outperforming VMamba by 3.70% mIoU under uneven illumination, 3.55% mIoU under complex backgrounds, and 5.10% mIoU under coexisting symptoms. Cross-dataset validation on Apple leaf diseases further proves its generalization with 3.39% mIoU and 3.84% Dice improvements over U-Mamba, while maintaining a competitive inference speed of 30 FPS. Qualitative results also show clearer boundaries, fewer missed small lesions, and more stable predictions in complex field scenarios. Full article

Hidden adulteration of stale berries in dried goji berry batches is difficult to detect by manual inspection or batch-level quality assessment. This study developed a high-throughput method for kernel-level spoilage adulteration quantification in dried goji berries. It addressed three practical challenges in the image processing of densely arranged dried-fruits, including scalable label generation for deep learning segmentation without pixel-level manual annotation, separation of densely touching small berries, and full-size quality level distribution map reconstruction. SAM-assisted pseudo-label generation combined with multi-scale image cropping was used to overcome the limitation of manual pixel-level annotation, while YOLO-based instance segmentation was further employed for efficient berry localization in dense scenes. The freshness labels of segmented single berries were assigned by a statistical RGB-HSV grading rule. Specifically, adaptive multi-scale image cropping for segmentation was applied to improve local separability of berries under dense adhesion and occlusion conditions. The crop-level segmentation and grading outputs were subsequently reconstructed into the original image coordinate system to generate complete quality distribution maps. Results showed that YOLO models trained based on the pseudo-labels achieved a precision of 0.953, a recall of 0.951, an mAP₅₀ of 0.960, and an mAP_50-95 of 0.846. The full-size grading map reconstruction method produced a mean duplicate-suppression rate of 4.31%. In the full freshness-grading test dataset, 4850 berries were detected, including 449 stale berries. The mean absolute counting error was 1.61%. The proposed framework reduces manual annotation requirements while enabling berry-level freshness classification and quantitative stale-berry proportion estimation, providing objective information for dried fruit quality screening and adulteration control. Full article

The remote sensing scene classification (RSSC) task plays a pivotal role in Earth observation missions, yet its progress remains constrained by the scarcity of high-quality labeled imagery. This article introduces a self-supervised learning (SSL) paradigm to address this challenge. First, for pseudo-label construction, a large set of long-interval satellite revisit imagery is collected and processed with pixel-level registration. The SIFT inliers retained during registration serve as saliency priors to guide asymmetric masking across views. This produces positive pairs that preserve global scene consistency while introducing controlled object-level ambiguities. Second, we propose a progressive layer-wise contrastive learning framework (MTC-Net) that couples the pseudo-label with the network’s representational hierarchy, forming a curriculum from local texture robustness to global semantic invariance. A dual-attention module with spatial–channel branches is further embedded to recalibrate intermediate features. The learning paradigm encourages the model to perform cross-view contextual reasoning rather than relying on pixel-wise correspondences. Experiments on three widely used datasets demonstrate that MTC-Net achieves competitive classification accuracy under limited-label settings, while ablation and visualization studies validate the effectiveness of establishing scene-level invariance through multi-temporal contrastive alignment. Full article

Surface vegetation dynamics in mining areas are characterized by complex non-linear processes associated with anthropogenic disturbance and ecological restoration. Existing remote sensing approaches often face limitations in balancing temporal interpretability and the characterization of long-term vegetation trajectories at regional scales. To address this issue, this study proposes a trajectory-guided weakly supervised framework that integrates parameterized curve fitting with deep temporal learning for mining vegetation monitoring. Based on the characteristic “extraction–reclamation” cycle, six representative vegetation trajectory patterns were pre-defined to describe different stages of degradation and restoration. Long-term NDVI trajectories (1990–2023) derived from Landsat time-series data were modeled using linear and parameterized Sigmoid functions to automatically generate high-quality supervision samples and temporal transition labels. These trajectory-constrained samples were subsequently incorporated into a multi-task BiLSTM-Attention network to simultaneously perform pixel-level change classification and turning-point regression. Applied to the mining clusters of the Dongting Lake Basin, China, the proposed framework achieved an overall classification accuracy of 86.64% (Kappa = 0.83), while the temporal prediction error remained within two years. Results revealed that 28.66% of the 61.20 km² of significantly degraded mining land has undergone effective ecological restoration, with restoration activities increasing sharply between 2012 and 2014 in response to regional environmental policies. By coupling ecological trajectory modeling with weakly supervised temporal learning, this study offers a promising approach for large-scale mining restoration monitoring and ecological assessment. Full article

Supraglacial lake depth is a key variable for quantifying surface meltwater storage and assessing ice-shelf stability, yet spatially continuous and reliable bathymetric information remains difficult to obtain in polar regions because in situ measurements are scarce and optical imagery cannot directly provide water depth. This study develops an integrated framework for supraglacial lake identification and bathymetry retrieval by combining ICESat-2 ATL03 photon-counting lidar data with Sentinel-2 multispectral imagery. ICESat-2 lake photons were used to constrain lake-region extraction from Sentinel-2 imagery, and the photon-derived along-track depths were corrected for scattering and refraction before being converted into Sentinel-2 pixel-level depth labels. Based on these labels, four retrieval models were constructed and evaluated, including an empirical model, CatBoost, a convolutional neural network (CNN), and a residual dense network (RDN). CatBoost generated initial depth estimates, while CNN and RDN further incorporated the CatBoost-derived depth prior and Sentinel-2 multispectral features for pixel-level depth prediction. Experiments over four investigated supraglacial lakes showed that RDN achieved the best average performance across the investigated lakes, with mean R², RMSE, and MAE values of 0.927, 0.187 m, and 0.144 m, respectively. For the investigated lakes, the integration of ICESat-2 and Sentinel-2 extended discrete along-track reference-depth observations to spatially continuous bathymetry maps. Because the training and validation samples were obtained from different spatial blocks within the same four lake scenes, the reported performance primarily reflects within-lake spatial generalization under the investigated conditions, and transferability to unseen lakes remains to be evaluated. These maps may provide inputs for future lake-volume estimation and ice-shelf hydrological analyses, while their applicability to lakes with different morphological and optical conditions requires further evaluation. Full article