UAV-based greenhouse rose monitoring requires accurate detection of rose targets under substantial scale variation, foliage occlusion, background similarity, illumination changes, and motion-induced degradation while maintaining lightweight inference for practical real-time edge deployment. This study proposes RoseScaleWaveNet (RSWNet), a YOLO26n-based lightweight detector that coordinates scale, spatial context, and frequency information through three complementary modules: C3MKIR for multi-receptive-field feature extraction, SEASCPP for position-aware multi-scale contextual modeling, and MWAdaLF for frequency-decoupled cross-level feature fusion. Experiments on a multi-source greenhouse rose dataset annotated with two petal-opening classes show that RSWNet achieves 94.13% precision, 91.95% recall, 95.51% mAP@50, and 59.22% mAP@50:95 with only 2.30 M parameters and an inference speed of 185 FPS. Compared with YOLO26n, mAP@50 and mAP@50:95 are improved by 5.50 and 4.70 percentage points, respectively, while model complexity is slightly reduced. The proposed modules also yield an average mAP@50 gain of 4.24 percentage points when transferred to four representative YOLO architectures. Cross-domain evaluation on six external flower-related datasets further demonstrates consistent improvements under both zero-shot evaluation and dataset-specific retraining. Controlled tests under blur, noise, illumination variation, motion blur, and scale changes characterize robustness to the evaluated image perturbations. Separate benchmarks on desktop and laptop GPUs, CPU-only hardware, and Jetson Orin Nano quantify inference efficiency. These results support the potential of RSWNet for UAV-assisted greenhouse rose detection within the evaluated conditions.

The accurate detection of weld defects is critical for ensuring structural integrity in industrial applications. However, traditional methods often face limitations in data scarcity, and manual annotation costs. This paper integrates a DDPM for data augmentation with a semi-supervised teacher-student architecture for efficient model training. To address data scarcity, DDPM is employed to synthesize high-quality weld defect images. A semi-supervised learning strategy, enhanced by Exponential Moving Average (EMA) and a Crop-Paste data augmentation technique, is utilized to leverage both labeled and unlabeled data effectively. And our method is evaluated against mainstream generative and segmentation models. Results demonstrate that our approach achieves superior performance, with the DDPM generating high-fidelity images (PSNR: 27.95 dB, SSIM: 77.38) and the final segmentation model attaining a mIoU of 88.47, significantly outperforming comparative benchmarks. This work provides a robust and efficient solution for automated weld defect inspection.

Strategic financial markets can admit multiple approximate Nash equilibria with materially different liquidation and price-impact regimes. Returning one equilibrium cannot exclude a more stressed alternative, yet complete enclosure becomes expensive when the profitable deviator changes inside a search region. We propose EquiPartition, a learning-augmented branch-and-bound method based on player-adaptive deviation certificates. The certificate combines institution-specific unilateral-deviation atoms through state-dependent Bernstein weights, while an outward-rounded verifier alone authorizes exclusion. Inconclusive proposals undergo bounded repair or conservative subdivision, so prediction quality affects efficiency but not the complete $\epsilon$-Nash outer-enclosure guarantee. On the controlled and multi-institution synthetic one-period fire-sale markets, the finite recorded-reference audit retained all recorded high-precision reference-equilibrium components and found no accepted exclusion intersecting them. Relative to Nash-B&B, Predict+Repair reduced proof-tree nodes by 83.3% and 83.9%, unresolved volume by 75.8% and 69.0%, and mean capped runtime by 40.5% and 20.2%; it completed 113 of 120 multi-institution instances versus 103 for Nash-B&B. The same finite recorded-reference audit retained all recorded crisis components under the financial stress shifts. Runtime decomposition showed that amortized proposals avoided repeated neutral-start optimization and offset repair and verification costs. EquiPartition reduces proof fragmentation while preserving verifier-based completeness.

Accurate and efficient clothing detection remains challenging because visually similar garment categories can share appearance cues, while background interference, scale variation, and spatial overlap weaken category-discriminative features. To address these issues, this study proposes a lightweight clothing detector based on YOLO26n with hierarchical discriminative feature refinement. The framework progressively enhances garment representations at the channel, semantic, and multi-scale spatial levels. Specifically, the C3k2-based Channel Feature Enhancement Module (C3k2_CFEM) strengthens cross-channel interaction and adaptively balances local details with contextual information. The C2-based Gated Semantic Attention module (C2GSA) selectively retains task-relevant high-level responses while suppressing redundant contextual interference. The Multi-Scale Dual-Stream Attention Module (MSDSAM) combines complementary foreground–background attention with efficient multi-scale feature interaction. A multi-source clothing dataset containing 11,970 images and 21,504 annotated garment instances across seven categories was used for evaluation. The proposed model achieves 91.6% Precision, 86.2% Recall, 92.6% mAP₅₀, and 74.7% mAP_50:95, improving the YOLO26n baseline by 6.6 and 5.8 percentage points in mAP₅₀ and mAP_50:95, respectively. These results are obtained with 2.39 M parameters, 5.4 billion floating-point operations (GFLOPs), and an inference speed of 166 frames per second (FPS) on an NVIDIA RTX 4090 at $640\times 640$ input resolution and batch size 1. Additional evaluations report the model’s behavior across clothing categories, target scales, overlap levels, image perturbations, hardware platforms, and external clothing datasets. Under the reported protocols, the framework improves detection accuracy with limited increases in parameters and computational cost.

Camera-based visual sensing for classroom behavior detection remains challenging because fine-grained behaviors often differ only in subtle posture and hand–object cues, while dense seating, occlusion, scale variation, and background interference further weaken discriminative visual evidence. To address these issues, this study proposes the Multi-scale Context, Hierarchical Integration, and Memory Enhancement Detector (MCHIME-Det), a lightweight, multi-scale context and memory-enhanced detector built on YOLO26n. The framework introduces three complementary components: C3k2PMM combines multi-scale local modeling with direction-aware selective state-space scanning; CGHFF performs channel-guided hierarchical feature interaction across adjacent feature levels; and C2MECM uses input-dependent memory experts to compensate for ambiguous high-level representations. For evaluation, two public classroom behavior datasets were combined into a corpus containing 11,879 images and 71,657 annotated instances across three categories: hand-raising, reading, and writing. On the curated image-level split, MCHIME-Det achieves 76.3% precision, 74.9% recall, 82.1% mAP₅₀, and 61.4% mAP_50:95, improving the YOLO26n baseline by 5.6 and 4.4 percentage points in mAP₅₀ and mAP_50:95, respectively. These gains are achieved with only 2.36 M parameters and 5.2 GFLOPs, while maintaining 182 FPS on an RTX 4090 and 17.1 FPS on a Jetson Orin Nano edge device. Ablation, cross-architecture, scale-, density-, and overlap-conditioned analyses, cross-source evaluation, image-corruption robustness tests, and cross-hardware benchmarks characterize model performance on the curated benchmark, three sampled external datasets, specified synthetic corruptions, and tested hardware platforms. These results support hierarchical local–global refinement and memory-assisted compensation as effective strategies for lightweight camera-based visual sensing under the evaluated image datasets and inference settings.