Search Results (1,409)

Smoke detection is a practical approach for early identification of wildfires and mitigating hazards that affect ecosystems, infrastructure, property, and the community. The existing deep learning (DL) object detection methods (e.g., Detection Transformer (DETR)) have demonstrated significant potential for early awareness of these events. However, their precision is influenced by the low visual salience of smoke and the reliability of the annotation, and collecting real-world and reliable datasets with precise annotations is a labor-intensive and time-consuming process. To address this challenge, we propose a weakly supervised Transformer-based approach with a teacher–student architecture designed explicitly for smoke detection while reducing the need for extensive labeling efforts. In the proposed approach, an expert model serves as the teacher, guiding the student model to learn from a variety of data annotations, including bounding boxes, point labels, and unlabeled images. This adaptability reduces the dependency on exhaustive manual annotation. The proposed approach integrates a Deformable-DETR backbone with a modified loss function to enhance the detection pipeline by improving spatial reasoning, supporting multi-scale feature learning, and facilitating a deeper understanding of the global context. The experimental results demonstrate performance comparable to, and in some cases exceeding, that of fully supervised models, including DETR and YOLOv8. Moreover, this study expands the existing datasets to offer a more comprehensive resource for the research community. Full article

The automated inspection of civil infrastructure with Unmanned Aerial Vehicles (UAVs) is hampered by the challenge of accurately segmenting multi-damage in high-resolution imagery. While foundational models like the Segment Anything Model (SAM) offer data-efficient segmentation, their effectiveness is constrained by prompting strategies, especially for geometrically complex defects. This paper presents a comprehensive comparative analysis of deep learning strategies to identify an optimal deep learning pipeline for segmenting cracks, efflorescences, and exposed rebars. It systematically evaluates three distinct end-to-end segmentation frameworks: the native output of a YOLO11 model; the Segment Anything Model (SAM), prompted by bounding boxes; and SAM, guided by a point-prompting mechanism derived from the detector’s probability map. Based on these findings, a final, optimized hybrid pipeline is proposed: for linear cracks, the native segmentation output of the SAHI-trained YOLO model is used, while for efflorescence and exposed rebar, the model’s bounding boxes are used to prompt SAM for a refined segmentation. This class-specific strategy yielded a final mean Average Precision (mAP50) of 0.593, with class-specific Intersection over Union (IoU) scores of 0.495 (cracks), 0.331 (efflorescence), and 0.205 (exposed rebar). The results establish that the future of automated inspection lies in intelligent frameworks that leverage the respective strengths of specialized detectors and powerful foundation models in a context-aware manner. Full article

A tracking method based on adaptive morphological correlation and neural predictive models is presented. The morphological correlation filters are optimized according to the aggregated binary dissimilarity-to-matching ratio criterion and are adapted online to appearance variations of the target across frames. Morphological correlation filtering enables reliable detection and accurate localization of the target in the scene. Furthermore, trained neural models predict the target’s expected location in subsequent frames and estimate its bounding box from the correlation response. Effective stages for drift correction and tracker reinitialization are also proposed. Performance evaluation results for the proposed tracking method on four image datasets are presented and discussed using objective measures of detection rate (DR), location accuracy in terms of normalized location error (NLE), and region-of-support estimation in terms of intersection over union (IoU). The results indicate a maximum average performance of 90.1% in DR, 0.754 in IoU, and 0.004 in NLE on a single dataset, and 83.9%, 0.694, and 0.015, respectively, across all four datasets. In addition, the results obtained with the proposed tracking method are compared with those of five widely used correlation filter-based trackers. The results show that the suggested morphological-correlation filtering, combined with trained neural models, generalizes well across diverse tracking conditions. Full article

Unmanned aerial vehicle (UAV) remote sensing is critical in assessing post-earthquake building damage. However, intelligent disaster assessment via remote sensing faces formidable challenges from complex backgrounds, substantial scale variations in targets, and diverse spatial disaster dynamics. To address these issues, we propose UEBNet, a high-precision post-earthquake building instance segmentation model that systematically enhances damage recognition by integrating three key modules. Firstly, the Depthwise Separable Convolutional Block Attention Module suppresses background noise that visually resembles damaged structures. This is achieved by expanding the receptive field using multi-scale pooling and dilated convolutions. Secondly, the Multi-feature Fusion Module generates scale-robust feature representations for damaged buildings with significant size differences by processing feature streams from different receptive fields in parallel. Finally, the Adaptive Multi-Scale Interaction Module accurately reconstructs the irregular contours of damaged buildings through an advanced feature alignment mechanism. Extensive experiments were conducted using UAV imagery collected after the Ms 6.8 earthquake in Tingri County, Tibet Autonomous Region, China, on 7 January 2025, and the Ms 6.2 earthquake in Jishishan County, Gansu Province, China, on 18 December 2023. Results indicate that UEBNet enhances segmentation mean Average Precision (${mAP}^{seg}$) and bounding box mean Average Precision (${mAP}^{box}$) by 3.09% and 2.20%, respectively, with equivalent improvements of 2.65% in F1-score and 1.54% in overall accuracy, outperforming state-of-the-art instance segmentation models. These results demonstrate the effectiveness and reliability of UEBNet in accurately segmenting earthquake-damaged buildings in complex post-disaster scenarios, offering valuable support for emergency response and disaster relief. Full article

Object detection in degraded environments such as low-light and nighttime conditions remains a challenging task, as conventional computer vision techniques often fail to achieve high precision and robust performance. With the increasing adoption of deep learning, this paper aims to enhance object detection under such adverse conditions by proposing an improved version of YOLOv11, named DRF-YOLO (Degradation-Robust and Feature-enhanced YOLO). The proposed framework incorporates three innovative components: (1) a lightweight Cross Stage Partial Multi-Scale Edge Enhancement (CSP-MSEE) module that combines multi-scale feature extraction with edge enhancement to strengthen feature representation; (2) a Focal Modulation attention mechanism that improves the network’s responsiveness to target regions and contextual information; and (3) a self-developed Dynamic Interaction Head (DIH) that enhances detection accuracy and spatial adaptability for small objects. In addition, a lightweight unsupervised image enhancement algorithm, Zero-DCE (Zero-Reference Deep Curve Estimation), is introduced prior to training to improve image contrast and detail, and Generalized Intersection over Union (GIoU) is employed as the bounding box regression loss. To evaluate the effectiveness of DRF-YOLO, experiments are conducted on two representative low-light datasets: ExDark and the nighttime subset of BDD100K, which include images of vehicles, pedestrians, and other road objects. Results show that DRF-YOLO achieves improvements of 3.4% and 2.3% in mAP@0.5 compared with the original YOLOv11, demonstrating enhanced robustness and accuracy in degraded environments while maintaining lightweight efficiency. Full article

Background: The automated analysis of intestinal organoids in microscopy images are essential for high-throughput morphological studies, enabling precision and scalability. Traditional manual analysis is time-consuming and subject to observer bias, whereas Machine Learning (ML) approaches have recently demonstrated superior performance. Purpose: This study aims to evaluate YOLO (You Only Look Once) for organoid segmentation and classification, comparing its standalone performance with a hybrid pipeline that integrates DL-based feature extraction and ML classifiers. Methods: The dataset, consisting of 840 light microscopy images and over 23,000 annotated intestinal organoids, was divided into training (756 images) and validation (84 images) sets. Organoids were categorized into four morphological classes: cystic non-budding organoids (Org0), early organoids (Org1), late organoids (Org3), and Spheroids (Sph). YOLO version 10 (YOLOv10) was trained as a segmenter-classifier for the detection and classification of organoids. Performance metrics for YOLOv10 as a standalone model included Average Precision (AP), mean AP at 50% overlap (mAP50), and confusion matrix evaluated on the validation set. In the hybrid pipeline, trained YOLOv10 segmented bounding boxes, and features extracted from these regions using YOLOv10 and ResNet50 were classified with ML algorithms, including Logistic Regression, Naive Bayes, K-Nearest Neighbors (KNN), Random Forest, eXtreme Gradient Boosting (XGBoost), and Multi-Layer Perceptrons (MLP). The performance of these classifiers was assessed using the Receiver Operating Characteristic (ROC) curve and its corresponding Area Under the Curve (AUC), precision, F1 score, and confusion matrix metrics. Principal Component Analysis (PCA) was applied to reduce feature dimensionality while retaining 95% of cumulative variance. To optimize the classification results, an ensemble approach based on AUC-weighted probability fusion was implemented to combine predictions across classifiers. Results: YOLOv10 as a standalone model achieved an overall mAP50 of 0.845, with high AP across all four classes (range 0.797–0.901). In the hybrid pipeline, features extracted with ResNet50 outperformed those extracted with YOLO, with multiple classifiers achieving AUC scores ranging from 0.71 to 0.98 on the validation set. Among all classifiers, Logistic Regression emerged as the best-performing model, achieving the highest AUC scores across multiple classes (range 0.93–0.98). Feature selection using PCA did not improve classification performance. The AUC-weighted ensemble method further enhanced performance, leveraging the strengths of multiple classifiers to optimize prediction, as demonstrated by improved ROC-AUC scores across all organoid classes (range 0.92–0.98). Conclusions: This study demonstrates the effectiveness of YOLOv10 as a standalone model and the robustness of hybrid pipelines combining ResNet50 feature extraction and ML classifiers. Logistic Regression emerged as the best-performing classifier, achieving the highest ROC-AUC across multiple classes. This approach ensures reproducible, automated, and precise morphological analysis, with significant potential for high-throughput organoid studies and live imaging applications. Full article

Background/Objectives: Endoscopic ultrasound (EUS) is crucial for lymph node (LN) characterization, playing a key role in oncological staging and treatment guidance. EUS criteria for predicting malignancy are imprecise, and histologic diagnosis may have limitations. This multicenter study aimed to evaluate the effectiveness of a novel artificial intelligence (AI)–based system in predicting LN malignancy from EUS images. Methods: This multicenter study included EUS images from nine centers. Lesions were labeled (“malignant” or “benign”) and delimited with bounding boxes. Definitive diagnoses were based on cytology/biopsy or surgical specimens and, if negative, a minimum six-month clinical follow-up. A convolutional neural network (CNN) was developed using the YOLO (You Only Look Once) architecture, incorporating both detection and classification modules. Results: A total of 59,992 images from 82 EUS procedures were analyzed. The CNN distinguished malignant from benign lymph nodes with a sensitivity of 98.8% (95% CI: 98.5–99.2%), specificity of 99.0% (95% CI: 98.3–99.7%), and precision of 99.0% (95% CI: 98.4–99.7%). The negative and positive predictive values for malignancy were 98.8% and 99.0%, respectively. Overall diagnostic accuracy was 98.3% (95% CI: 97.6–99.1%). Conclusions: This is the first study evaluating the performance of deep learning systems for LN assessment using EUS imaging. Our AI-powered imaging model shows excellent detection and classification capabilities, emphasizing its potential to provide a valuable tool to refine LN evaluation with EUS, ultimately supporting more tailored, efficient patient care. Full article

The rapid expansion of the goose farming industry creates a growing need for real-time flock counting and individual-level behavior monitoring. To meet this challenge, this study proposes an improved YOLOv8-based model, termed DAEF-YOLO (DualConv-augmented C2f, ADown down-sampling, Efficient Channel Attention integrated into SPPF, and FocalerIoU regression loss), designed for simultaneous recognition of Sanhua goose individuals and their diverse behaviors. The model incorporates three targeted architectural improvements: (1) a C2f-Dual module that enhances multi-scale feature extraction and fusion, (2) ECA embedded in the SPPF module to refine channel interaction with minimal parameter cost and (3) an ADown down-sampling module that preserves cross-channel information continuity while reducing information loss. Additionally, the adoption of the FocalerIoU loss function enhances bounding-box regression accuracy in complex detection scenarios. Experimental results demonstrate that DAEF-YOLO surpasses YOLOv5s, YOLOv7-Tiny, YOLOv7, YOLOv9s, and YOLOv10s in both accuracy and computational efficiency. Compared with YOLOv8s, DAEF-YOLO achieved a 4.56% increase in precision, 6.37% in recall, 5.50% in F1-score, and 4.59% in mAP@0.5, reaching 94.65%, 92.17%, 93.39%, and 96.10%, respectively. A generalizable classification strategy is further introduced by adding a complementary “Other” category to include behaviors beyond predefined classes. This approach ensures complete recognition coverage and demonstrates strong transferability for multi-task detection across species and environments. Ablation studies indicated that mAP@0.5 remained consistent (~96.1%), while mAP@0.5:0.95 improved in the absence of the “Other” class (75.68% vs. 69.82%). Despite this trade-off, incorporating the “Other” category ensures annotation completeness and more robust multi-task behavior recognition under real-world variability. Full article

This paper presents Rain-Cloud Condensation Optimizer (RCCO), a nature-inspired metaheuristic that maps cloud microphysics to population-based search. Candidate solutions (“droplets”) evolve under a dual-attractor dynamic toward both a global leader and a rank-weighted cloud core, with time-decaying coefficients that progressively shift emphasis from exploration to exploitation. Diversity is preserved via domain-aware coalescence and opposition-based mirroring sampled within the coordinate-wise band defined by two parents. Rare heavy-tailed “turbulence gusts” (Cauchy perturbations) enable long jumps, while a wrap-and-reflect scheme enforces feasibility near the bounds. A sine-map initializer improves early coverage with negligible overhead. RCCO exposes a small hyperparameter set, and its per-iteration time and memory scale linearly with population size and problem dimension. RCOO has been compared with 21 state-of-the-art optimizers, over the CEC 2022 benchmark suite, where it achieves competitive to superior accuracy and stability, and achieves the top results over eight functions, including in high-dimensional regimes. We further demonstrate constrained, real-world effectiveness on five structural engineering problems—cantilever stepped beam, pressure vessel, planetary gear train, ten-bar planar truss, and three-bar truss. These results suggest that a hydrology-inspired search framework, coupled with simple state-dependent schedules, yields a robust, low-tuning optimizer for black-box, nonconvex problems. Full article

To address the challenges of poor model adaptability and high computational complexity in complex orchard environments characterized by variable lighting, severe occlusion, and dense fruit clusters, an enhanced kiwifruit detection network, Kiwi-YOLO, is proposed based on YOLOv8. Firstly, replacing the main network with the MobileViTv1 module reduces computational load and parameters, thus enhancing inference efficiency for mobile deployment. Secondly, incorporating BiFPN into the model’s neck as a replacement for PANet improves feature distinguishability between background regions and target instances. Additionally, incorporating MCA module promotes cross-dimensional feature interactions, strengthening model robustness and generalization performance. Finally, the MPDIoU loss function is adopted to minimize bounding box vertex distances, mitigating detection box distortion caused by sample heterogeneity while accelerating convergence and enhancing localization accuracy. Experimental results indicate that the enhanced model achieves improvements of 2.1%, 1.5% and 0.3% in precision, recall, and mAP, respectively, over the baseline YOLOv8, while reducing parameters (Params) and computational complexity (GFLOPs) by 19.71 million and 2.8 billion operations. Moreover, it surpasses other comparative models in performance. Furthermore, in experiments detecting kiwifruit targets under complex lighting and occlusion conditions, the Kiwi-YOLO model demonstrated excellent adaptability and robustness. Its strong environmental adaptability provides technical guidance for advancing the practical application of unmanned intelligent kiwifruit harvesting. Full article

This work presents an Internet of Things (IoT) node designed for low-power agrifood chain tracking in remote areas, where long-range terrestrial communication is either unavailable or severely limited. The novelty of this study lies in the development and characterization of an IoT node prototype that leverages direct-to-satellite connectivity through a geostationary Earth orbit (GEO) satellite, using long-range frequency-hopping spread spectrum (LR-FHSS) modulation in the licensed S-band. The prototype integrates a microcontroller unit that manages both the radio modem and a suite of sensors, enclosed in a plastic box suitable for field deployment. Characterization in an anechoic chamber demonstrated a maximum effective isotropic radiated power (EIRP) of $27.5$ dBm, sufficient to establish a reliable satellite link. The onboard sensors provide global positioning as well as measurements of acceleration, temperature, humidity, and solar radiation intensity. Prototype performance was assessed in two representative scenarios: stationary and mobile. Regarding energy consumption, the average charge drained by the radio modem per transmission cycle was measured to be 356 mC. With a battery pack composed of four 2500 mAh NiMH cells, the estimated upper bound on the number of transmitted packets is approximately 25,000. Full article

The intelligent transformation of power systems necessitates robust insulator condition detection to ensure grid safety. Existing methods, primarily reliant on manual inspection or conventional image processing, suffer significantly degraded target identification and detection efficiency under extreme weather conditions such as heavy snowfall. To address this challenge, this paper proposes an enhanced YOLO11s detection framework integrated with image restoration technology, specifically targeting insulator defect identification in snowy environments. First, data augmentation and a FocalNet-based snow removal algorithm effectively enhance image resolution under snow conditions, enabling the construction of a high-quality training dataset. Next, the model architecture incorporates a dynamic snake convolution module to strengthen the perception of tubular structural features, while the MPDIoU loss function optimizes bounding box localization accuracy and recall. Comparative experiments demonstrate that the optimized framework significantly improves overall detection performance under complex weather compared to the baseline model. Furthermore, it exhibits clear advantages over current mainstream detection models. This approach provides a novel technical solution for monitoring power equipment conditions in extreme weather, offering significant practical value for ensuring reliable grid operation. Full article

Accurate pedestrian detection from a robotic perspective has become increasingly critical, especially in complex environments such as crowded and high-density populations. Existing methods have low accuracy due to multi-scale pedestrians and dense occlusion in complex environments. To address the above drawbacks, a dense pedestrian detection network architecture based on YOLOv8n (DPDN-YOLOv8) was introduced for complex environments. The network aims to improve robots’ pedestrian detection in complex environments. Firstly, the C2f modules in the backbone network are replaced with C2f_ODConv modules integrating omni-dimensional dynamic convolution (ODConv) to enable the model’s multi-dimensional feature focusing on detected targets. Secondly, the up-sampling operator Content-Aware Reassembly of Features (CARAFE) is presented to replace the Up-Sample module to reduce the loss of the up-sampling information. Then, the Adaptive Spatial Feature Fusion detector head with four detector heads (ASFF-4) was introduced to enhance the system’s ability to detect small targets. Finally, to accelerate the convergence of the network, the Focaler-Shape-IoU is utilized to become the bounding box regression loss function. The experimental results show that, compared with YOLOv8n, the mAP@0.5 of DPDN-YOLOv8 increases from 80.5% to 85.6%. Although model parameters increase from $3\times {10}^6$ to $5.2\times {10}^6$, it can still meet requirements for deployment on mobile devices. Full article

Student engagement analysis is closely linked with learning outcomes, and its precise identification paves the way for targeted instruction and personalized learning. Current student engagement methods, reliant on either head pose estimation with facial landmarks or eye-trackers, are hardly generalized to authentic classroom teaching environments with high occlusion and non-intrusive requirements. Based on empirical observations that student body orientation and head pose exhibit a high degree of consistency in classroom settings, we propose a novel student engagement analysis algorithm incorporating human body orientation estimation. To better suit classroom settings, we develop a one-stage and end-to-end trainable framework for multi-person body orientation estimation, named JointBDOE. The proposed JointBDOE integrates human bounding box prediction and body orientation into a unified embedding space, enabling the simultaneous and precise estimation of human positions and orientations in multi-person scenarios. Extensive experimental results using the MEBOW dataset demonstrate the superior performance of JointBDOE over the state-of-the-art methods, with an MAE reduced to 10.63° and orientation accuracy exceeding 91% at 22.5°. With the more challenging reconstructed MEBOW dataset, JointBDOE maintains strong robustness with an MAE of 16.07° and an orientation accuracy of 88.3% at 30°. Further analysis of classroom teaching videos validates the reliability and practical value of body orientation as a robust metric for engagement assessment. This research showcases the potential of artificial intelligence in intelligent classroom analysis and provides an extensible solution for body orientation estimation technology in related fields, advancing the practical application of intelligent educational tools. Full article

Accurate underground utility mapping remains a critical yet complex task in Ground Penetrating Radar (GPR) interpretation, essential to avoiding costly and dangerous excavation errors. This study presents a novel deep learning-based pipeline for 3D reconstruction of buried linear utilities from high-resolution GPR B-scan data. Three state-of-the-art models—YOLOv8, YOLOv11, and Mask R-CNN—were employed for both bounding box and keypoint detection of hyperbolic reflections, using a real-world GPR dataset. On the test set, Mask R-CNN achieved the highest keypoint F1-score (0.822) and bounding box F1-score (0.867), outperforming the YOLO models. Detected summit points were clustered using a 3D DBSCAN algorithm to approximate the spatial trajectories of buried utilities. RANSAC-based line fitting was then applied to each cluster, yielding an average RMSE of 0.06 across all fitted 3D paths. The key innovation of this hybrid model lies in its use of real-world data (avoiding synthetic augmentation), direct summit point detection (beyond bounding box analysis), and a geometric 3D reconstruction pipeline. This approach addresses key limitations in prior studies, including poor generalizability to complex real-world scenarios and the reliance on full 3D data volumes. Our method offers a more practical and scalable solution for subsurface utility mapping in real-world settings. Full article