Search Results (201)

Nowadays, object detection algorithms are widely used in various scenarios. However, there are further small object detection requirements in some special scenarios. Due to the problems related to small objects, such as their less available features, unbalanced samples, higher positioning accuracy requirements, and fewer data sets, a small object detection algorithm is more complex than a general object detection algorithm. The detection effect of the model for small objects is not ideal. Therefore, this paper takes YOLOXs as the benchmark network and enhances the feature information on small objects by improving the network’s structure so as to improve the detection effect of the model for small objects. This specific research is presented as follows: Aiming at the problem of a neck network based on an FPN and its variants being prone to information loss in the feature fusion of non-adjacent layers, this paper proposes a feature fusion and distribution module, which replaces the information transmission path, from deep to shallow, in the neck network of YOLOXs. This method first fuses and extracts the feature layers used by the backbone network for prediction to obtain global feature information containing multiple-size objects. Then, the global feature information is distributed to each prediction branch to ensure that the high-level semantic and fine-grained information are more efficiently integrated so as to help the model effectively learn the discriminative information on small objects and classify them correctly. Finally, after testing on the VisDrone2021 dataset, which corresponds to a standard image size of 1080p (1920 × 1080), the resolution of each image is high and the video frame rate contained in the dataset is usually 30 frames/second (fps), with a high resolution in time, it can be used to detect objects of various sizes and for dynamic object detection tasks. And when we integrated the module into a YOLOXs network (named the FE-YOLO network) with the three improvement points of the feature layer, channel number, and maximum pool, the mAP and APs were increased by 1.0% and 0.8%, respectively. Compared with YOLOV5m, YOLOV7-Tiny, FCOS, and other advanced models, it can obtain the best performance. Full article

This study utilizes a novel albedo retrieval framework combining radiative transfer modeling with scientific machine learning (RTM-SciML) to investigate sea ice dynamics in the Sea of Okhotsk. By validating albedo estimates derived from the MODIS sensor against in situ pyranometer measurements near the Hokkaido coast, we achieved a robust Pearson coefficient of 0.86 and an RMSE of 0.089 for all sea ice types, with even higher correlations for specific surfaces like snow-covered ice (Pearson-r = 0.89) and meltwater/open water (Pearson-r = 0.90). This confirms the framework’s efficacy across varying surface conditions. Cross-sensor comparisons between MODIS and the Second-Generation Global Imager (SGLI) further demonstrated its consistency, achieving an overall Pearson-r of 0.883 and RMSE of 0.036. Integrating these albedo estimates with sea ice concentration data from the Advanced Microwave Scanning Radiometer 2 (AMSR-2), we analyzed the complex role of the Sea of Okhotsk’s polynya systems and ice interactions in regional climate processes. Our results highlight the synergistic advantage of pairing optical sensors, like MODIS and SGLI, with microwave sensors, offering a more comprehensive understanding of evolving sea ice conditions and paving the way for future climate and cryosphere studies. Full article

Deep connectionist models are characterized by many neurons grouped together in many successive layers. As a result, their data classifications are difficult to understand. We present two novel algorithms which explain the responses of several black-box machine learning models. The first is Fidex, which is local and thus applied to a single sample. The second, called FidexGlo, is global and uses Fidex. Both algorithms generate explanations by means of propositional rules. In our framework, the discriminative boundaries are parallel to the input variables and their location is precisely determined. Fidex is a heuristic algorithm that, at each step, establishes where the best hyperplane is that has increased fidelity the most. The algorithmic complexity of Fidex is proportional to the maximum number of steps, the number of possible hyperplanes, which is finite, and the number of samples. We first used FidexGlo with ensembles and support vector machines (SVMs) to show that its performance on three benchmark problems is competitive in terms of complexity, fidelity and accuracy. The most challenging part was then to apply it to convolutional neural networks. We achieved this with three classification problems based on images. We obtained accurate results and described the characteristics of the rules generated, as well as several examples of explanations illustrated with their corresponding images. To the best of our knowledge, this is one of the few works showing a global rule extraction technique applied to both ensembles, SVMs and deep neural networks. Full article

Brain tumors, both primary and metastatic, represent a significant global health burden due to their high incidence, mortality, and the severe neurological deficits they frequently cause. Gliomas, especially high-grade gliomas (HGGs), rank among the most aggressive and lethal neoplasms, with only modest gains in long-term survival despite extensive molecular research and established standard therapies. In neurosurgical practice, maximizing the extent of safe resection is a principal strategy for improving clinical outcomes. Yet, the infiltrative nature of gliomas often complicates the accurate delineation of tumor margins. Confocal laser endomicroscopy (CLE), originally introduced in gastroenterology, has recently gained prominence in neuro-oncology by enabling real-time, high-resolution cellular imaging during surgery. This technique allows for intraoperative tumor characterization and reduces dependence on time-consuming frozen-section analyses. Recent technological advances, including device miniaturization and second-generation CLE systems, have substantially improved image quality and diagnostic utility. Furthermore, integration with deep learning algorithms and telepathology platforms fosters automated image interpretation and remote expert consultations, thereby accelerating surgical decision making and enhancing diagnostic consistency. Future work should address remaining challenges, such as mitigating motion artifacts, refining training protocols, and broadening the range of applicable fluorescent probes, to solidify CLE’s role as a critical intraoperative adjunct in neurosurgical oncology. Full article

Effective management to conserve marine environments requires up-to-date information on the location, distribution, and extent of major benthic habitats. Remote sensing is a key tool for such assessments, enabling consistent, repeated measurements over large areas. There is particular interest in using freely available satellite images such as from the Copernicus Sentinel-2 series for accessible repeat assessments. In this study, an area of 438 km² of the northern Red Sea coastline, adjacent to the NEOM development was mapped using Sentinel-2 imagery. A hierarchical Random Forest classification method was used, where the initial level classified pixels into a geomorphological class, followed by a second level of benthic cover classification. Uncrewed Aerial Vehicle (UAV) surveys were carried out in 12 locations in the NEOM area to collect field data on benthic cover for training and validation. The overall accuracy of the geomorphic and benthic classifications was 84.15% and 72.97%, respectively. Approximately 12% (26.26 km²) of the shallow Red Sea study area was classified as coral or dense algae and 16% (36.12 km²) was classified as rubble. These reef environments offer crucial ecosystem services and are believed to be internationally important as a global warming refugium. Seagrass meadows, covering an estimated 29.17 km² of the study area, play a regionally significant role in carbon sequestration and are estimated to store 200 tonnes of carbon annually, emphasising the importance of their conservation for meeting the environmental goals of the NEOM megaproject. This is the first map of this region generated using Sentinel-2 data and demonstrates the feasibility of using an open source and reproducible methodology for monitoring coastal habitats in the region. The use of training data derived from UAV imagery provides a low-cost and time-efficient alternative to traditional methods of boat or snorkel surveys for covering large areas in remote sites. Full article

The Münchberg Gneiss Complex (Central European Variscides, Germany) is separated by a deep-seated lineamentary fault zone, the Franconian Lineamentary Fault Zone, from its Mesozoic foreland. The study area offers insight into a great variety of landforms created by fluvial and mass wasting processes together with their bedrocks, covering the full range from unmetamorphosed sediments to high-grade regionally metamorphic rocks. It renders the region an ideal place to conduct a study of compositional and numerical geomorphology and their landscape-forming indices and parameters. The landforms under consideration are sculpted out of the bedrocks (erosional landforms) and overlain by depositional landforms which are discussed by means of numerical landform indices (LFIs), all of which are coined for the first time in the current paper. They are designed to be suitable for applied geosciences such as extractive/economic geology as well as environmental geology. The erosional landform series are subdivided into three categories: (1) The landscape roughness indices, e.g., VeSi_val (vertical sinuosity—valley of landform series) and the VaSlAn_alti (variation in slope angle altitude), which are used for a first order classification of landscapes into relief generations. The second order classification LFIs are devoted to the material properties of the landforms’ bedrocks, such as the rock strength (VeSi_lith) and the bedrock anisotropy (VaSlAn_norm). The third order scheme describes the hydrography as to its vertical changes by the inclination of the talweg and the different types of knickpoints (IncTal_lith/grad) and horizontal sinuosity (HoSi_lith/grad). The study area is subjected to a tripartite zonation into the headwater zone, synonymous with the paleoplain which undergoes some dissection at its edge, the step-fault plain representative of the track zone which undergoes widespread fluvial piracy, and the foreland plains which act as an intermediate sedimentary trap named the deposition zone. The area can be described in space and time with these landform indices reflecting fluvial and mass wasting processes operative in four different stages (around 17 Ma, 6 to 4 Ma, <1.7 Ma, and <0.4 Ma). The various groups of LFIs are a function of landscape maturity (pre-mature, mature, and super-mature). The depositional landforms are numerically defined in the same way and only differ from each other by their subscripts. Their set of LFIs is a mirror image of the composition of depositional landforms in relation to their grain size. The leading part of the acronym, such as QuantSan_heav and QuantGrav_lith, refers to the process of quantification, the second part to the grain size, such as sand and gravel, and the subscript to the material, such as heavy minerals or lithological fragments. The three numerical indices applicable to depositional landforms are a direct measurement of the hydrodynamic and gravity-driven conditions of the fluvial and mass wasting processes using granulometry, grain morphology, and situmetry (clast orientation). Together with the previous compositional indices, the latter directly translate into the provenance analysis which can be used for environmental analyses and as a tool for mineral exploration. It creates a network between numerical geomorphology, geomorphometry, and the E&E issue disciplines (economic/extractive geology vs. environmental geology). The linguistics of the LFIs adopted in this publication are designed so as to be open for individual amendments by the reader. An easy adaptation to different landform suites worldwide, irrespective of their climatic conditions, geodynamic setting, and age of formation, is feasible due to the use of a software and a database available on a global basis. Full article

Breast cancer ranks as the second most prevalent cancer globally and is the most frequently diagnosed cancer among women; therefore, early, automated, and precise detection is essential. Most AI-based techniques for breast cancer detection are complex and have high computational costs. Hence, to overcome this challenge, we have presented the innovative LightweightUNet hybrid deep learning (DL) classifier for the accurate classification of breast cancer. The proposed model boasts a low computational cost due to its smaller number of layers in its architecture, and its adaptive nature stems from its use of depth-wise separable convolution. We have employed a multimodal approach to validate the model’s performance, using 13,000 images from two distinct modalities: mammogram imaging (MGI) and ultrasound imaging (USI). We collected the multimodal imaging datasets from seven different sources, including the benchmark datasets DDSM, MIAS, INbreast, BrEaST, BUSI, Thammasat, and HMSS. Since the datasets are from various sources, we have resized them to the uniform size of 256 × 256 pixels and normalized them using the Box-Cox transformation technique. Since the USI dataset is smaller, we have applied the StyleGAN3 model to generate 10,000 synthetic ultrasound images. In this work, we have performed two separate experiments: the first on a real dataset without augmentation and the second on a real + GAN-augmented dataset using our proposed method. During the experiments, we used a 5-fold cross-validation method, and our proposed model obtained good results on the real dataset (87.16% precision, 86.87% recall, 86.84% F1-score, and 86.87% accuracy) without adding any extra data. Similarly, the second experiment provides better performance on the real + GAN-augmented dataset (96.36% precision, 96.35% recall, 96.35% F1-score, and 96.35% accuracy). This multimodal approach, which utilizes LightweightUNet, enhances the performance by 9.20% in precision, 9.48% in recall, 9.51% in F1-score, and a 9.48% increase in accuracy on the combined dataset. The LightweightUNet model we proposed works very well thanks to a creative network design, adding fake images to the data, and a multimodal training method. These results show that the model has a lot of potential for use in clinical settings. Full article

The Global Change Observation Mission-Climate (GCOM-C) satellite, launched in December 2017, is equipped with the Second-generation Global Imager (SGLI) sensor, featuring a moderate spatial resolution of 250 m and 19 spectral bands, including the unique 380 nm band. After six years in orbit, a comprehensive evaluation of SGLI products and their temporal consistency is needed. Remote sensing reflectance ($R_{rs}$) is the primary product for monitoring water quality, forming the basis for deriving key oceanic constituents such as chlorophyll-a (Chla) and total suspended matter (TSM). The Japan Aerospace Exploration Agency (JAXA) provides $R_{rs}$ products through two platforms, G-Portal and JASMES, each employing different atmospheric correction methodologies and assumptions. This study aims to evaluate the SGLI full-resolution $R_{rs}$ products from G-Portal and JASMES at regional scales (Japan and East Asia) and assess G-Portal $R_{rs}$ products globally between January 2018 and December 2023. The evaluation employs in situ matchups from NASA’s Aerosol Robotic Network-Ocean Color (AERONET-OC) and cruise measurements. We also assess the retrieval accuracy of two water quality indices, Chla and TSM. The AERONET-OC data analysis reveals that JASMES systematically underestimates $R_{rs}$ values at shorter wavelengths, particularly at 412 nm. While the $R_{rs}$ accuracy at 412 nm is relatively low, G-Portal’s $R_{rs}$ products perform better than JASMES at shorter wavelengths, showing lower errors and stronger correlations with AERONET-OC data. Both G-Portal and JASMES show lower agreement with AERONET-OC and cruise datasets at shorter wavelengths but demonstrate improved agreement at longer wavelengths (530 nm, 565 nm, and 670 nm). JASMES generates approximately 12% more matchup data points than G-Portal, likely due to G-Portal’s stricter atmospheric correction thresholds that exclude pixels with high reflectance. In situ measurements indicate that G-Portal provides better overall agreement, particularly at lower $R_{rs}$ magnitudes and Chla concentrations below 5 mg/m³. This evaluation underscores the complexities and challenges of atmospheric correction, particularly in optically complex coastal waters (Case 2 waters), which may require tailored atmospheric correction methods different from the standard approach. The assessment of temporal consistency and seasonal variations in $R_{rs}$ data shows that both platforms effectively capture interannual trends and maintain temporal stability, particularly from the 490 nm band onward, underscoring the potential of SGLI data for long-term monitoring of coastal and oceanic environments. Full article

Buildings are fundamental elements of human environments, and detecting changes in them is crucial for land cover studies, urban expansion monitoring, and the detection of illegal construction activities. Existing methods primarily focus on pixel-level differences in bi-temporal remote sensing imagery. However, pseudo-changes, such as variations in non-building areas caused by differences in illumination, seasonal changes, and other factors, pose significant challenges for reliable building change detection. To address these issues, we propose a novel object-level contrastive-learning-based multi-branch network (OCL-Net) for detecting building changes by integrating bi-temporal remote sensing images. First, we design a multi-head decoder to separately extract more distinguishable building change features and auxiliary semantic features from bi-temporal images, effectively leveraging building-specific priors. Second, an object-level contrastive learning loss is designed and jointly optimized with a pixel-level similarity loss to ensure the global consistency of buildings. Finally, an attention-based discriminative feature generation and fusion block is designed to enhance the representation of multi-scale change features. We validate the effectiveness of the proposed method through comparative experiments on the publicly available WHU-CD and S2Looking datasets. Our approach achieves IoU values of 88.54% and 51.94%, respectively, surpassing state-of-the-art methods for building change detection. Full article

Falls are defined by the World Health Organization (WHO) as incidents in which an individual unintentionally falls to the ground or a lower level. Falls represent a serious public health issue, ranking as the second leading cause of death from unintentional injuries, following traffic accidents. While fall prevention is crucial, prompt intervention after a fall is equally necessary. Delayed responses can result in severe complications, reduced recovery potential, and a negative impact on quality of life. This study focuses on detecting fall situations using image-based methods. The fall images utilized in this research were created by combining three open-source datasets to enhance generalization and adaptability across diverse scenarios. Because falls must be detected promptly, the YOLO (You Only Look Once) network, known for its effectiveness in real-time detection, was applied. To better capture the complex body structures and interactions with the floor during a fall, two key techniques were integrated. First, a global attention module (GAM) based on the Convolutional Block Attention Module (CBAM) was employed to improve detection performance. Second, a Transformer-based Swin Transformer module was added to effectively learn global spatial information and enable a more detailed analysis of body movements. This study prioritized minimizing missed fall detections (false negatives, FN) as the key performance metric, since undetected falls pose greater risks than false detections. The proposed Fall Detection YOLO (FD-YOLO) network, developed by integrating the Swin Transformer and GAM into YOLOv9, achieved a high [email protected] score of 0.982 and recorded only 134 missed fall incidents, demonstrating optimal performance. When implemented in environments equipped with standard camera systems, the proposed FD-YOLO network is expected to enable real-time fall detection and prompt post-fall responses. This technology has the potential to significantly improve public health and safety by preventing fall-related injuries and facilitating rapid interventions. Full article

Hyperspectral compressed imaging is a novel imaging detection technology based on compressed sensing theory that can quickly acquire spectral information of terrestrial objects in a single exposure. It combines reconstruction algorithms to recover hyperspectral data from low-dimensional measurement images. However, hyperspectral images from different scenes often exhibit high-frequency data sparsity and existing deep reconstruction algorithms struggle to establish accurate mapping models, leading to issues with detail loss in the reconstruction results. To address this issue, we propose a hyperspectral reconstruction method based on global gradient information and local low-rank priors. First, to improve the prior model’s efficiency in utilizing information of different frequencies, we design a gradient sampling strategy and training framework based on decision trees, leveraging changes in the loss function gradient information to enhance the model’s predictive capability for data of varying frequencies. Second, utilizing the local low-rank prior characteristics of the representative coefficient matrix, we develop a sparse sensing denoising module to effectively improve the local smoothness of point predictions. Finally, by establishing a regularization term for the reconstruction process based on the semantic similarity between the denoised results and prior spectral data, we ensure spatial consistency and spectral fidelity in the reconstruction results. Experimental results indicate that the proposed method achieves better detail recovery across different scenes, demonstrates improved generalization performance for reconstructing information of various frequencies, and yields higher reconstruction quality. Full article

Traditional multimodal contrastive learning brings text and its corresponding image closer together as a positive pair, where the text typically consists of fixed sentence structures or specific descriptive statements, and the image features are generally global features (with some fine-grained work using local features). Similar to unimodal self-supervised contrastive learning, this approach can be seen as enforcing a strict identity constraint in a multimodal context. However, due to the inherent complexity of remote sensing images, which cannot be easily described in a single sentence, and the fact that remote sensing images contain rich ancillary information beyond just object features, this strict identity constraint may be insufficient. To fully leverage the characteristics of remote sensing images, we propose a multimodal contrastive learning method for remote sensing image feature extraction, based on positive sample tripartite relaxation, where the model is relaxed in three aspects. The first aspect of relaxation involves both the text and image inputs. By introducing learnable parameters in the language and image branches, instead of relying on fixed sentence structures and fixed image features, the network can achieve a more flexible description of remote sensing images in text and extract ancillary information from the image features, thereby relaxing the input constraints. Second relaxation is achieved through multimodal alignment of various features. By aligning semantic information with the corresponding semantic regions in the images, the method allows for the relaxation of local image features under semantic constraints. This approach addresses the issue of selecting image patches in unimodal settings, where there is no semantic constraint. The proposed method for remote sensing image feature extraction has been validated on four datasets. On the PatternNet dataset, it achieved a 91.1% accuracy with just one-shot. Full article

An unescapable trend of hyperspectral image (HSI) has been toward classification with high accuracy and splendid performance. In recent years, Transformers have made remarkable progress in the HSI classification task. However, Transformer-based methods still encounter two main challenges. First, they concentrate on extracting spectral information and are incapable of using spatial information to a great extent. Second, they lack the utilization of multiscale features and do not sufficiently combine the advantages of the Transformer’s global feature extraction and multiscale feature extraction. To tackle these challenges, this article proposes a new solution named the hybrid-scale dual-domain Transformer with crisscrossed interaction (HSD²Former) for HSI classification. HSD²Former consists of three functional modules: dual-dimension multiscale convolutional embedding (D²MSCE), mixed domainFormer (MDFormer), and pyramid scale fusion block (PSFB). D²MSCE supersedes conventional patch embedding to generate spectral and spatial tokens at different scales, effectively enriching the diversity of spectral-spatial features. MDFormer is designed to facilitate self-enhancement and information interaction between the spectral domain and spatial domain, alleviating the heterogeneity of the spatial domain and spectral domain. PSFB introduces a straightforward fusion manner to achieve advanced semantic information for classification. Extensive experiments conducted on four datasets demonstrate the robustness and significance of HSD²Former. The classification evaluation indicators of OA, AA, and Kappa on four datasets almost exceed 98%, reaching state-of-the-art performance. Full article

Accurate 3D object detection is essential for autonomous driving, yet traditional LiDAR models often struggle with sparse point clouds. We propose perspective-aware hierarchical vision transformer-based LiDAR-camera fusion (PLC-Fusion) for 3D object detection to address this. This efficient, multi-modal 3D object detection framework integrates LiDAR and camera data for improved performance. First, our method enhances LiDAR data by projecting them onto a 2D plane, enabling the extraction of object perspective features from a probability map via the Object Perspective Sampling (OPS) module. It incorporates a lightweight perspective detector, consisting of interconnected 2D and monocular 3D sub-networks, to extract image features and generate object perspective proposals by predicting and refining top-scored 3D candidates. Second, it leverages two independent transformers—CamViT for 2D image features and LidViT for 3D point cloud features. These ViT-based representations are fused via the Cross-Fusion module for hierarchical and deep representation learning, improving performance and computational efficiency. These mechanisms enhance the utilization of semantic features in a region of interest (ROI) to obtain more representative point features, leading to a more effective fusion of information from both LiDAR and camera sources. PLC-Fusion outperforms existing methods, achieving a mean average precision (mAP) of 83.52% and 90.37% for 3D and BEV detection, respectively. Moreover, PLC-Fusion maintains a competitive inference time of 0.18 s. Our model addresses computational bottlenecks by eliminating the need for dense BEV searches and global attention mechanisms while improving detection range and precision. Full article

Background: Determinants and motives related to food selection have evolved in a globalized and changing world. The traditional and useful Food Choice Questionnaire (FCQ), created in 1995, needs to be updated, adapted to new scenarios, and validated. Objectives: This study aimed to: (1) assess face validity (FV) of the original 36-item FCQ, (2) generate an Updated-FCQ (U-FCQ) and assess its content validity (CV) (instrument suitability), and (3) evaluate its construct validity and reliability in a Spanish-speaking population from Mexico. Methods: FV involved a panel of nutrition professionals (NPs) rating the original items’ clarity, relevance, specificity, and representativeness. A literature review process updated the FCQ by adding new items. CV with a second NP panel allowed calculating content validity ratio (CVR). Construct validation was performed via exploratory and confirmatory factor analysis (EFA-CFA). Internal consistency through Cronbach’s alpha (CA) and test–retest reliability via intra-class correlation (ICC) were assessed. Results: The FV (n = 8) resulted in the modification of 11 original items. The literature review added 36 new items (15 from previous adaptations and 21 original items). The CV (n = 13) identified nine items (non-acceptable CVR), prompting reformulation of seven and removal of two. The NPs’ feedback added six new items. The EFA-CFA (n = 788) developed a 75-item U-FCQ with eight dimensions: sensory appeal, mood, health and nutritional content, price, food identity, environmental and wildlife awareness, convenience, and image management. CA ranged from 0.74–0.97 (good–excellent) and ICC from 0.51–0.78 (moderate–good). Conclusions: This study provides a useful instrument for the assessment of food choices and lays the groundwork for future cross-cultural comparisons, expanding its applicability in wider settings. Full article