Search Results (983)

Two-stage oriented detectors achieve high localization accuracy in synthetic aperture radar (SAR) ship detection, but their large backbones, feature pyramids, proposal modules, and heavy region of interest (RoI) heads hinder deployment. Existing lightweight SAR ship detectors typically use one-stage frameworks that lack proposal-level refinement for precise rotated localization. This paper presents a dual-teacher knowledge distillation and pruning (DTKDP) framework for lightweight oriented SAR ship detection. DTKDP introduces learnable gates into convolutional, normalization, and linear layers to prune convolutional channels and RoI-head neurons. Rotated proposal alignment (RPA) distills teacher and student predictions in a shared teacher-generated rotated proposal space, while a dual-teacher scheme combines classification and regression guidance from a homogeneous main teacher with complementary classification cues from a heterogeneous auxiliary teacher. Experiments on the SAR Ship Detection Dataset (SSDD) and Rotated Ship Detection Dataset in SAR Images (RSDD-SAR) show that DTKDP reduces the parameters of Oriented Region-based Convolutional Neural Network (Oriented R-CNN) and RoI Transformer equipped with ResNet-50 backbones by 87.5–91.8% and their floating-point operations (FLOPs) by 75.6–79.9%. In terms of average precision (AP) and mean average precision (mAP), the resulting Oriented R-CNN-slim and RoI Transformer-slim retain accuracy close to their full-scale counterparts. Relative changes across AP${}_{50}$, AP${}_{75}$, mAP${}_{50:75}$, and mAP${}_{50:95}$ range from a 2.38% decrease to a 0.65% improvement. Compared with RTMDet-tiny, they improve all four metrics on both datasets by 0.52–27.55% and consistently surpass representative distillation methods, demonstrating a favorable accuracy–efficiency trade-off. Full article

Against China’s dual-carbon goals, carbon emissions and pollution transfer across supply chains hinder systemic low-carbon transformation; the prior literature on ESG and corporate environmental investment mainly focuses on individual firm-level effects, ignoring the unique lead goose governance function of core chain leader enterprises in supply chain networks and their internal transmission mechanisms. Drawing on stakeholder theory, this paper constructs a collaborative green supply chain governance framework led by chain leaders to fill the above research gaps. Based on a sample of Shanghai and Shenzhen A-share listed firms from 2011 to 2023, we identify chain leaders by combining official industrial chain leader lists and total asset threshold standards, with baseline ESG data from Wind and alternative Bloomberg ESG scores for robustness. Two-stage least squares instrumental variable regression addresses endogeneity, while omitted variable sensitivity analysis, indicator replacement and stepwise high-dimensional fixed effects ensure reliable empirical conclusions. The results provide evidence of a significant lead goose spillover effect: each one-unit improvement in chain leaders’ ESG performance is associated with an 8.33% increase in upstream suppliers’ environmental investment at the 1% significance level and downstream clients’ environmental investment by 3.54% at the 5% significance level, yet non-leader enterprises generate no meaningful spillover impacts, and the effect is stronger for upstream partners. Mechanism tests support two core mediating channels: chain leaders’ ESG performance stimulates supply-chain green investment by fostering environmental sensitivity salience and cutting inter-firm transaction costs. Heterogeneity analysis shows the spillover effect is amplified in polluting industries and highly concentrated supply chains. This research extends the emerging literature on supply chain ESG spillovers by documenting the lead goose effect of formally identified chain leaders on partners’ actual environmental investment, distinguishing directional asymmetry, and unveiling dual mediating mechanisms. Full article

Large language model agents invoke external tools through the Model Context Protocol to perform actions beyond text generation. Once a background thread is dispatched, even the latest Model Context Protocol Tasks extension exposes only coarse-grained, self-reported task states, so a thread that deadlocks or crashes silently remains reported as working and invisible to the client, which is not acceptable in some industrial applications that need long monitoring. We present ReliHarness, a self-learning dual-channel execution-layer reliability harness framework that instead probes each worker thread at a lower level through two independent checks: heartbeat timeout detection and operating system-level thread liveness detection. The proposed framework contributes a five-state lifecycle machine, dual-channel detection, bi-layer evidence logging, a four-stage cascade pipeline, and a self-learning command-matching module. Experiments on a Phytium S5000C and Ascend 310P platform achieve 100% detection coverage for the tested crash and deadlock scenarios with zero false positives. The self-learning module expands the knowledge base and lifts matching accuracy from 74.5% to 90.2% across 102 real-world test queries. The per-task overhead is 0.004 ms, and regex responses are 670 to 63,300 times faster than large language model inference. The architecture is application and logic decoupled and deployable in Model Context Protocol-compatible systems. Full article

This paper examines the dynamic relationship between innovation, total factor productivity (TFP), and economic growth in Ghana using annual data for the period 1965–2021. Although Ghana has recorded relatively strong economic growth, concerns remain regarding the sustainability of this performance in the absence of consistent productivity improvements. The study combines growth accounting techniques with time-series econometric methods, including the autoregressive distributed lag–unrestricted error correction model (ARDL–UECM), vector error correction modelling (VECM), Granger causality tests, and two-stage least squares estimation. The results provide robust evidence of a stable long-run equilibrium relationship among innovation, productivity, and output. Innovation exerts a positive and statistically significant effect on economic growth, primarily through productivity-enhancing channels, while TFP emerges as the dominant long-run driver of growth. Short-run dynamics reveal feedback effects between innovation, productivity, and economic growth. However, growth accounting results indicate substantial volatility in TFP growth, suggesting that Ghana’s expansion has been driven largely by factor accumulation rather than sustained efficiency gains. The findings offer policy-relevant insights for productivity-centred growth strategies in Sub-Saharan Africa. Full article

Multivariate time series forecasting requires modeling both hierarchical temporal dynamics and complex inter-variable dependencies, a dual requirement that often degrades predictive performance and incurs high computational costs in standard Transformer architectures. Unlike current channel-independent models that ignore vital cross-variable synergies, or dense-attention frameworks that suffer from quadratic computational noise, our approach extracts structurally sparse dependencies. To address these specific limitations, this study introduces the FCA-Transformer. The proposed framework integrates a Feature Pyramid Network (FPN) to isolate macroscopic trends from high-frequency localized fluctuations via hierarchical downsampling. Concurrently, a structured Transformer-based Cross-Attention (TCA) mechanism employs Dimensional Segmentation with Weighting (DSW) and a Two-Stage Attention (TSA) layer to map topological variable interactions, effectively extracting robust cross-variable pathways and mitigating distributional noise. Extensive empirical evaluations across three real-world multivariate benchmarks (ETTh1, Electricity, and Exchange Rate) demonstrate that the FCA-Transformer achieves an average reduction of up to 4.39% in MSE and 5.11% in MAE compared to leading baselines. These findings indicate that the proposed architecture successfully reconciles multi-scale feature extraction with lightweight dependency modeling, enhancing structural generalization and providing a scalable framework for real-time temporal analysis in complex industrial environments. Full article

Using survey data from 1025 farm households in Anhui, Henan, Hebei, and Shandong Provinces, this study examines whether farmers’ participation in digitally enabled agricultural activities is associated with farm expansion in China’s major grain-producing areas. Digital economy participation is measured across four stages of the agricultural value chain: procurement, production, supply and marketing, and finance. Farm expansion is assessed using two analytically parallel outcomes: entry into the farmland rental market and log operated farmland area. A double/debiased machine learning framework flexibly adjusts for observed household, local, and county-level covariates and potentially nonlinear relationships. Digital economy participation is positively associated with both outcomes, and the results remain qualitatively robust across alternative index weights, parametric benchmarks, county fixed effects, learning algorithms, and supplementary instrumental-variable specifications. Channel analyses show that digital participation is positively associated with information acquisition, market access, and credit accessibility; the sequential results are consistent with information acquisition supporting market access and credit accessibility. Heterogeneity analysis further shows that household capabilities are more closely related to differences in farmland rental market entry, whereas local labor and mechanization conditions are more relevant to realized operational expansion. These findings clarify the role and limitations of agricultural digital participation in household land adjustment and provide evidence for improving digital and land-related services in China’s major grain-producing regions. Full article

This article presents a low-cost, high-voltage excitation circuit (EC) for capacitive ultra-sonic transducers (CUTs) based on a dual-path architecture. The proposed design comprises two independent AC excitation channels (AC-branch) that share a regulated DC-bias voltage (DC-branch). The circuit was developed to satisfy a fundamental operational requirement of CUTs: simultaneous application of a static bias voltage and a time-varying drive voltage. Because the electrostatic force depends nonlinearly on the applied voltage, efficient first-harmonic actuation requires the superposition of DC and AC voltage components. To reach this objective, the circuit and the transducer must be treated as a coupled electrical, electrostatic, mechanical, and acoustic system. In the proposed implementation, the DC-branch uses a TL494-PWM controller, a TIP50 switching transistor, a step-up transformer, and a rectifier-filter stage to generate the high-voltage bias of up to 200 V_DC. Each AC-channel employs an LM3886TF amplifier followed by a 1:15 step-up transformer, enabling the generation of excitation signals of up 180 V_peak. A key feature of the proposed architecture is the electrical independence of the two AC-channels, which allows for distinct excitation frequencies with minimal mutual interference. Experimental validation, performed with and without ultrasonic loads, demonstrates the relation between excitation conditions and the acoustic performance of the CUTs. Full article

A long-standing account of retinal direction selectivity states that asymmetric inhibition suppresses responses to motion in the null direction. We use this idea as a computational prior in the Directional Inhibition Network (DI-Net), a two-stage model for direction classification from a before/after image pair. The first stage is a deterministic, parameter-free, anti-coincidence encoder that produces eight spatial maps of local directional evidence. The second is a compact convolutional network that combines this evidence into a global direction estimate. Because only the second stage is learned, the intermediate code remains directly inspectable. Experiments use object-conditioned pairs derived from DIS5K, with controlled translations and corruption applied only at test time. For one-pixel motion, DI-Net achieves 0.993 Accuracy on clean pairs and 0.747 under 10% corruption, close to Lucas–Kanade in the same evaluation. Fixed-channel voting reduces clean Accuracy to 0.713, whereas a convolutional network trained directly on the image pair degrades much more sharply under corruption. A multi-step version of the encoder also improves direction classification for displacements from 1 to 16 pixels compared with a parameter-matched one-step control. Tests on selected DAVIS-derived pairs show no statistically resolved difference between DI-Net and the evaluated optical-flow baselines. Taken together, these results support DI-Net as an interpretable, retina-inspired computational model for controlled motion-direction tasks rather than as a physiological account of retinal processing. Full article

On-demand droplet formation is of crucial importance to the engineering applications of droplet microfluidics. This work presents an experimental investigation on active control of droplet formation using stacked piezoelectric actuators. Two stacked piezoelectric actuators are placed adjacent to the continuous phase channel, producing periodic excitations on the continuous phase flow. It is found that droplet formation greatly depends on the excitation frequency and voltage. Droplet formation synchronizes piezoelectric excitation at a small excitation frequency, i.e., droplet formation frequency equals excitation frequency and its subharmonics. Beyond a critical value of the excitation frequency, a neglected effect of excitation frequency on droplet generation is observed. The droplet generation frequency could be increased up to ~2.6 times that without excitation. The droplet generation frequency exhibits a stepwise increase with rising excitation voltage. When the droplet formation frequency equals the excitation frequency, droplet formation undergoes filling, necking, and pinching-off. When the droplet formation frequency is half of the excitation frequency, additional refilling and re-necking stages are observed. The regime diagram of the droplet formation frequency in the synchronization mode is presented. The scaling of the generated droplet length is deduced. Since periodic excitations are employed on the continuous phase flow, the proposed active control method could minimize the detrimental impact on biochemical reagents within droplets. Full article

Whether green-bond issuance is associated with shareholder value remains unclear in small, concentrated emerging markets, where signaling and information-asymmetry channels from developed-market studies may not operate the same way. This study examines short-term stock-market reactions and medium-term firm valuation associated with green-bond issuance among the full population of 20 eligible Thai listed issuers: 34 analytical issuance events from May 2019 to May 2026, and an annual panel of 20 firms observed from 2015 to 2025, comprising 220 repeated firm-year observations and 187 complete cases in the baseline regression. Event-study models find no statistically detectable average abnormal return around the issue date, robust across benchmark models and event windows; two-way fixed-effect panel regressions, likewise, find no statistically detectable average association with Tobin’s Q, with a confidence interval wide enough to admit economically meaningful effects in either direction. The matching and selection analyses remain inconclusive. The IV model has a weak first stage and is therefore reported only as a diagnostic. Because the design identifies within-firm variation among issuers, rather than a comparison with matched non-issuers, the results should be read as associations, rather than causal effects. The study provides census-based evidence that, in Thailand’s small and concentrated green-bond market, a green label may not yet carry detectable average value relevance for shareholders. Full article

This paper proposes Phase-PCANet, a new image representation method for fingerprint liveness detection. Phase-PCANet integrates both local and global phase features. The local phase, capturing fine-grained edges and textures, is extracted via short-time Fourier transform with singular value decomposition; the global phase, encoding the holistic structural layout, is obtained from a full-image Fourier transform. These two phase components are separately fed into an improved PCANet, which employs dual binary coding, i.e., scalar-based intra-channel coding and vector-similarity-based inter-channel coding, to preserve within-channel structures and cross-channel correlations. Multi-stage features from different PCA layers within each phase path are aggregated, and the resulting deep features from both paths are concatenated to form the final image representation. Experiments on the LivDet 2011, 2013, and 2015 databases verify the effectiveness of the proposed method. Full article

Transcutaneous electrical stimulation (TES) is a non-invasive technique that delivers electrical currents through the skin to modulate peripheral neural and muscular activity. However, conventional surface stimulation often exhibits limited spatial selectivity because current spreads across superficial and deep tissues. To investigate physical principles that may inform future transcutaneous applications without assuming anatomical or physiological fidelity, this study developed a finite-element framework based on an idealized multilayer cylindrical limb model. The model represented skin, adipose tissue, muscle, cortical bone, and bone marrow using concentric tissue domains parameterized with averaged forearm dimensions and literature-derived electrical properties. Two-dimensional and three-dimensional surface-electrode montages were evaluated by varying return-electrode positions, axial electrode spacing, and inter-channel current ratios. The simulations showed that electrode configuration influenced the location, volume, and compactness of high-maximum-envelope-modulation-amplitude (MEMA) regions. Within the investigated idealized geometry, intermediate return-electrode angles (approximately θ₂ = 130–150°) produced more centrally distributed intramuscular high-MEMA regions, whereas current-ratio modulation shifted the MEMA_max-defined field maximum under fixed electrodes. These results demonstrate the computational feasibility of regulating TI electric-field distributions in a controlled virtual model rather than physiological selectivity or clinical efficacy. Translation to actual transcutaneous stimulation will require staged validation using anatomically realistic models, physical phantoms, physiological experiments, and ultimately human studies. Full article

This study presents a systematic comparison of two dominant paradigms for improving rotation robustness in remote sensing semantic segmentation: rotation-invariant architectural modules (represented by Rotation-Invariant Channel Mapping, RICM) and random rotation data augmentation. Despite both approaches being widely adopted, a fair and side-by-side evaluation under unified experimental conditions remains lacking. Using U-Net as the baseline on the LoveDA dataset, we comprehensively evaluate six strategies: RICM alone, random rotation augmentation at probabilities $p=0.25$ and $p=0.5$, and their combinations. Performance is assessed under both discrete rotations (0°, 90°, 180°, 270°) and continuous angles (0°–180° with 15° intervals), supplemented by Rotation Consistency (RC) analysis, class-wise evaluation, and computational cost comparison. Cross-dataset validation on ISPRS Potsdam confirms generalizability. Our results demonstrate that: (1) random rotation augmentation consistently outperforms RICM, delivering substantial robustness gains with zero inference overhead; (2) augmentation probability controls a clear accuracy–robustness trade-off, with $p=0.25$ achieving the best balance (maintaining 0° mIoU at 74.8% while boosting 90° mIoU from 54.9% to 63.1%) and $p=0.5$ achieving near-invariance at the cost of reduced baseline accuracy; and (3) RICM offers only marginal benefits and becomes redundant when augmentation is applied. Analysis of the observed performance patterns suggests that early-stage RICM insertion may suppress useful orientation-specific features and that its rotation-ensemble averaging may be insufficient for dense prediction tasks. Overall, random rotation augmentation proves to be a simple, effective, and computationally efficient strategy for improving rotation robustness in remote sensing segmentation. Full article

Plant diseases, being a subject of interdisciplinary research, significantly reduce crop yield, quality, and economic returns, while the misidentification of pathogens often leads to ineffective treatments and may harm beneficial organisms and ecosystems. This work develops an approach for robust visual classification of plant diseases under limited and heterogeneous data based on multi-scale fractal texture descriptors integrated into a convolutional neural network. The proposed method employs wavelet transform modulus maxima to extract two complementary fractal characteristics, local fractal dimension and singularity spectrum width, from leaf images at several spatial scales. These descriptors form multi-channel fractal maps fed into a fractal attention module (FAM) inserted after the third stage of a ResNet-50 architecture. The FAM learns to emphasize spatial regions where fractal properties are most discriminative, while a parallel branch encodes global fractal statistics into an auxiliary vector combined with backbone features at the final classification layer. Experiments are conducted on a large heterogeneous collection of 11 public plant disease datasets under 5-shot, 50-shot, and full-scale training regimes. The fractal-augmented model raises classification accuracy from 57.06% to 67.73% on 5 shots and from 80.81% to 86.11% on 50 shots, red outperforming the plain ResNet-50 in these settings, converges within 1–2 epochs versus 25–40, and shows markedly better resilience to color distortions, random occlusions, and grayscale conversion in most cases. The generated attention maps provide spatially explicit explanations of the model’s decisions, increasing transparency for practical use. The proposed approach demonstrates that fractal analysis, embedded as a modulating signal inside a deep network, can serve as an efficient and interpretable inductive bias, which is particularly valuable under data scarcity and noisy agricultural imagery. Full article

Patients with early-stage myocardial infarction (MI) are at high risk of malignant ventricular arrhythmias, yet the cell-type-specific molecular landscape associated with post-infarction arrhythmogenesis has not been systematically characterized. This study integrates single-nucleus and spatial transcriptomics to define a cardiomyocyte subpopulation in early MI and dissect the NEAT1-centered regulatory network driving its ion channel remodeling. Single-nucleus transcriptomic data from post-MI human hearts were re-analyzed to identify a distinct subpopulation, termed arrhythmia-potential cardiomyocytes (aCMs), within the infarct border zone, characterized by pronounced ion channel remodeling. Gene co-expression network analysis revealed two modules highly associated with aCMs, in which NEAT1 correlated with the calcium channel gene CACNA1C and the LIM domain protein PDLIM5. All three genes were upregulated in hypoxic rat cardiomyocytes; siRNA-mediated knockdown confirmed that NEAT1 silencing downregulated CACNA1C and PDLIM5 expression, consistent with in silico knockout predictions. A ceRNA network further identified hsa-miR-204-5p/211-5p as a key mediator consistent with regulatory axis. These findings suggest that cardiomyocytes in the early MI border zone exhibit ion channel remodeling driven by elevated NEAT1, which may modulate CACNA1C and PDLIM5 through a microRNA-mediated ceRNA network, suggesting that targeting NEAT1 may warrant further investigation for preventing malignant arrhythmias in early-stage MI. Full article