Search Results (3,405)

Enterotoxigenic Escherichia coli K88 (ETEC-K88) is a primary causative agent of post-weaning diarrhea in piglets, yet most in vitro infection models evaluate epithelial injury at a limited number of endpoint time points, failing to capture the temporal dynamics of host–pathogen interactions. Here, we established a time-resolved ETEC-K88 challenge model in IPEC-J2 cells from 0 h to 48 h. The data showed that ETEC-K88 adhesion significantly increased after 4 h and peaked after 24 h, with a critical response transition at 4–8 h characterized by coordinated changes in mRNA expression of tryptophan metabolism, tight junction, aquaporins, and Solute Carrier Transporters (SLC). Additionally, we also evaluated the protective effects of L-tryptophan supplementation in IPEC-J2 with ETEC-K88 infection and found that its addition significantly restored the disrupted gene expression related to tryptophan metabolism, transporter channels, aquaporins, and cell cycle. Finally, two different mouse strains, C57BL/6J and BALB/c mice, were challenged with ETEC-K88 to assess strain- and segment-specific intestinal responses. Although overt diarrhea was not clearly induced in mice, the ETEC-K88 fimbria receptor genes were induced in both mice strains. Additionally, both mice strains exhibited obvious intestinal histomorphological abnormalities and showed the strain- and segment-specific expression of intestinal stem cell marker genes (Lgr5, SOX9), goblet cell marker gene TFF3, and aquaporin genes. In conclusion, we have defined a temporal epithelial response framework for ETEC-K88 infection in IPEC-J2 cells and mice, providing a theoretical basis for developing nutritional strategies against ETEC-associated intestinal dysfunction in pig production. Full article

Road traffic surveys require sensing systems that can be deployed rapidly without modifying the road surface or requiring a permanent power connection. This paper presents the design, embedded implementation, and evaluation of a low-power roadside magnetic sensor that performs vehicle-event detection and classification directly at the edge. The sensing node integrates two RM3100 three-axis magnetometers (PNI Sensor, Santa Rosa, CA, USA) with an NXP MK22FN512VLH12 microcontroller (NXP Semiconductors N.V., Eindhoven, The Netherlands) based on a 120 MHz Arm Cortex-M4F core with 512 kB Flash and 128 kB SRAM. Magnetic-field data are acquired at 250 Hz and processed locally using baseline removal, low-pass filtering, signal-energy calculation, and peak-based event detection. Detected magnetic signatures are classified using an integer-quantised one-dimensional convolutional neural network implemented directly on the microcontroller. The model processes four synchronised 512-sample channels representing the three magnetic-field axes and their combined signal energy. Model development was supported by approximately 50,000 annotated events obtained from 36 h of real-world traffic measurements at eight locations. The selected model achieved an overall classification accuracy of 91.1% for the considered operational categories. The implemented network requires 288,128 multiply–accumulate operations per inference, while its quantised weights and biases occupy approximately 23 kB of Flash memory. Complete three-axis event signatures are stored locally for subsequent verification, whereas only the timestamp and predicted vehicle category are transmitted through the wireless interface. Based on the capacity of the applied LiFePO₄ battery and the estimated consumption of the implemented hardware, the expected autonomous operating period is approximately 41 days. The results demonstrate the feasibility of integrating magnetic sensing, embedded signal processing, and Edge AI on a conventional resource-constrained Cortex-M4 platform for non-invasive road traffic monitoring. Full article

A one-dimensional Thue–Morse graphene photonic crystal (1D TMGPC) composed of alternating identical dielectric layers and graphene sheets is proposed, in which two distinct graphene chemical potentials are arranged according to a Thue–Morse quasi-periodic sequence. Using the transfer matrix method, we demonstrate that this structure effectively modulates terahertz waves and generates multiple abundant photonic bandgaps at both 20 K and 300 K. Notably, a novel splitting of low-frequency bandgaps produces two additional omnidirectional and polarization-insensitive bandgaps centered at approximately 1.45 THz and 1.95 THz. By analyzing the dispersion relations, reflection phase, photonic density of states, and electric field distributions, the boundary-driven modulation mechanism associated with the quasi-periodic chemical potential is elucidated. Furthermore, the proposed structure exhibits excellent multi-dimensional tunability. The bandgap properties can be dynamically tuned via the electrical control of graphene chemical potentials without altering the physical geometry. Structural tailoring provides an additional degree of freedom, as increasing the Thue–Morse sequence order induces passband splitting. Additionally, increasing the number of repeating periods yields comb-like multi-channel narrowband filtering responses. At a cryogenic temperature of 20 K, two distinct multi-channel narrowband comb filtering responses appear in the frequency ranges of 1.20–1.33 THz and 4.10–4.80 THz, with a minimum full width at half maximum (FWHM) of 1.10 GHz. At a room temperature of 300 K, the higher-frequency comb filtering response remains in the range of 4.10–4.80 THz, with a minimum FWHM of 5.70 GHz. Moreover, we evaluate the performance and stability of the structure when employed as filters and electro-optic switches, thereby providing useful insights for terahertz applications. With its simple geometry, abundant bandgaps, and flexible electro-structural tunability, the proposed 1D TMGPC is highly promising for broadband and electrically tunable terahertz devices. Full article

Building Information Modeling (BIM)-enabled performance assessment workflows for rural constructions remain relatively unexplored. This is even more important for buildings implementing process-oriented systems, such as airflow networks. This study presents a BIM-integrated framework for the early-stage design and performance assessment of rural constructions, with application to solar barn dryers and their ventilation systems through reduced-order airflow-network modeling. The proposed workflow combines parametric BIM-based geometry generation with lumped-parameter fluid-dynamic modeling to evaluate the influence of airflow-network topology on pressure losses, airflow distribution, fan power demand, and energy consumption. Nine BIM-generated design alternatives and ten geometric parameter sets were investigated under equivalent operating conditions. The airflow system was represented as a pressure-driven network including solar air panels, ducts, collectors, fan chambers, ventilation channels, and drying cells, accounting for both localized and distributed pressure losses. Results show that airflow-network geometry significantly affects system performance. Configurations characterized by more compact and aerodynamically efficient layouts reduced cumulative pressure losses by approximately 10–20% compared with less optimized solutions. More efficient designs enable reductions in required airflow rates of ~22% and in fan power demand of up to ~40% (≈11–18 kW). The most efficient configurations also exhibited lower annual energy consumption while maintaining the minimum overpressure required for effective hay drying. The study demonstrates how BIM environments can support physics-informed comparative evaluation of alternative ventilation layouts during the early design stage, extending BIM applications toward performance-oriented design and digital management of agricultural building systems. The proposed methodology provides a computationally efficient design-support framework that may also apply to other controlled-environment agricultural infrastructures governed by airflow-network dynamics. Full article

Objective: To address small bolt defect targets, complex background interference, and limited edge deployment in hilly and mountainous rail transport environments, a detection method balancing accuracy, lightweight design, and real-time performance was proposed. Methods: A track image dataset containing missing bolts, loose bolts, and missing nuts was constructed. Based on YOLOv11m, HFERBC3K2 was developed by replacing the standard bottleneck in C3K2 with a High-Frequency Enhancement Residual Block to strengthen edge, texture, and local structural feature extraction. A Spectral Enhanced Feed-Forward module was introduced into C2PSA to form SEFFNC2PSA, enhancing defect-related frequency components and suppressing background interference through adaptive frequency-domain modulation. The integrated model was named HFSE-YOLOv11. Channel-level structured pruning was then applied, and the model with a pruning ratio of 0.5 was named LHFSE-YOLOv11. Results: On the validation set, HFSE-YOLOv11 achieved 91.7% precision, 93.4% recall, 91.3% mAP@0.5, and 80.2% mAP@0.5:0.95, improving upon YOLOv11m by 3.6, 2.2, 1.2, and 3.1 percentage points, respectively. After pruning, LHFSE-YOLOv11 had 15.9 M parameters, 53.8 GFLOPs, and a 32.5 MB model size, representing reductions of 16.3%, 14.3%, and 11.7%, while mAP@0.5 and mAP@0.5:0.95 decreased by only 0.3 and 0.9 percentage points. On the independent test set, it achieved 91.7% precision, 93.4% recall, 91.0% mAP@0.5, 79.3% mAP@0.5:0.95, and 81.5 FPS, outperforming all compared models in the four detection metrics. Conclusion: LHFSE-YOLOv11 balances accuracy, efficiency, and model size, supporting deployment on vehicle-mounted inspection terminals and resource-constrained edge devices. Full article

This paper presents an adaptive event-triggered nonsingular fast terminal sliding-mode control (AETSMC) framework for quadrotor unmanned aerial vehicles subject to aerodynamic disturbances, actuator loss of effectiveness (LOE) of up to 60%, and intermittent denial-of-service (DoS) attacks. First, a nonsingular fast terminal sliding-mode (NFTSM) surface is constructed using fractional powers of the tracking error rather than fractional-order derivatives. This design ensures finite-time convergence while avoiding the singularity associated with conventional terminal sliding-mode schemes. Second, a smooth positive-semidefinite barrier function (Smooth-PSBF) is incorporated into the adaptive gain law. The resulting law provides only the compensation required to maintain the prescribed bound, thereby limiting gain overestimation and chattering. Third, a dual-mode event-triggering mechanism combines an exponentially decaying threshold with a zero-order hold. A positive lower bound on the inter-event interval is derived from the closed-loop dynamics, which excludes Zeno behaviour. Simulations under matched conditions show that the proposed method reduces the pitch-channel root-mean-square error by 79.4% and the integral squared error by 95.8% relative to the first reproduced baseline. In a separate 15-s communication experiment sampled at 1 kHz, the controller generated 128 transmissions instead of 15,000 periodic updates, corresponding to a 99.15% reduction. These results indicate that the proposed framework can improve fault-tolerant tracking while reducing communication demand under intermittent DoS attacks. Full article

Currently, insulation faults in the DC system of photovoltaic (PV) power stations are handled by a full shutdown strategy of inverters, and fault branch localization relies on manual inspection, resulting in low efficiency and poor accuracy, leading to prolonged unplanned outages and substantial power generation losses. This paper proposes an integrated solution combining high-precision insulation monitoring and intelligent fault line selection, which ensures the reliability of line selection criteria through improved measurement accuracy and achieves automatic fault isolation via optimized line selection strategies. The paper analyzes the mathematical essence of the ill-conditioned measurement equations of the traditional bridge method under severe single-pole grounding faults, establishes a dual-channel heteroscedastic noise model, and utilizes the inherent physical constraint that the sum of the positive and negative pole-to-ground voltages always equals the bus voltage to transform the ill-posed inverse problem into an equality-constrained optimal estimation problem, deriving an analytical solution in the sense of constrained least squares. A collaborative monitoring strategy of “balanced bridge monitoring first, unbalanced bridge precision measurement afterward” is proposed. An automatic fault line selection and isolation algorithm based on sequential branch switching is designed, which leverages the operational characteristic that PV systems allow short-term branch interruption, enabling automatic identification and isolation of faulty branches and automatic restoration of non-faulty branches without installing any leakage current sensors. Experimental results show that under severe fault conditions with a single-pole insulation resistance as low as 22 kΩ, the proposed method limits the error to within 5%; the proposed line selection strategy can complete identification and isolation of all faulty branches within at most two rounds of switching. Full article

Thermal transpiration enables a Knudsen pump to transport gas without moving components. An axial-integration formulation based on pre-computed transport coefficients from the linearized Shakhov kinetic model is applied to a single-stage unit comprising a circular microchannel and a circular macrochannel in series, with opposite wall-temperature gradients. The pressure-generation and gas-transport capabilities are characterized by the maximum pressure difference or thermomolecular pressure difference (TPD), the maximum mass flow rate, the equivalent TPD, the equivalent flow resistance, and the complete mass-flow-rate–pressure-difference characteristics. The principal quantitative calculations cover temperature differences ranging from 10 to 50 K, while the 75 and 100 K cases are retained only to assess the persistence of the calculated trends. The formulation reproduces benchmark experimental TPD data with a maximum absolute relative deviation of 12.5% and a mean absolute relative deviation of 6.7%, and shows excellent agreement with numerical data from the literature, with deviation below 1%. A decomposition of the microchannel and macrochannel contributions shows that a macrochannel contributing little to the total equivalent flow resistance may nevertheless produce appreciable reverse thermal transpiration. At the baseline condition of the study and for an inlet pressure $P_i=10 \mathrm{kPa}$, the macrochannel contributes only 0.37% of the total equivalent flow resistance but cancels 15.2% of the microchannel equivalent TPD. Over $P_i=1\text{–}50 \mathrm{kPa}$, the temperature-difference sensitivity of the TPD ranges from 0.93 to 1.05, whereas the microchannel-radius sensitivity varies from −0.41 to −1.62. For the maximum mass flow rate, the microchannel-radius sensitivity ranges from 2.06 to 2.56 and the microchannel-length sensitivity remains close to −1, while the macrochannel-length effect is negligible. Increasing the macrochannel radius improves both limiting outputs, i.e., TPD and maximum mass flow rate, but with progressively diminishing benefits from further enlargement of the macrochannel. These results provide a quantitative basis for preliminary dimension selection while explicitly identifying the limitations associated with linearization, finite channel length, fully developed flow, and neglected interface losses. Full article

Vehicle detection using drone-based imagery holds significant practical value in applications such as intelligent traffic management, urban situational awareness, and emergency response coordination. It is among the core technologies for achieving coordinated air–ground intelligent monitoring. However, when viewed from the high-altitude perspective of a drone, vehicle images suffer from issues such as complex backgrounds, blurriness, and low light, severely limiting the accuracy of model detection. Therefore, we propose a vehicle detection model for unmanned aerial vehicles based on YOLOv13 (VFD-YOLO). First, to address the issues of blurriness and low illumination in drone images, we designed a convolutional structure VFD-C3k2 specifically for vehicle feature extraction. It is a multi-scale feature-embedding structure based on the existing HLFD structure. We decomposed the vehicle image features into high-frequency details such as edges and textures, as well as low-frequency structural information such as global contours. Through differentiated processing, we enhanced the image restoration and detail extraction capabilities, and improved the adaptability of the model to different types of images. Afterward, to address the strong background interference from vehicle targets in drone imagery, we employed the CASAB channel and spatial attention module. This module enhances the weights of key vehicle feature channels through channel attention, focuses on the target area of the vehicle through spatial attention, effectively suppresses background noise, and strengthens the model’s ability to focus on and extract significant target features. The experimental results show that on the DroneVehicle dataset, the precision, recall rate, mAP@0.5 and mAP@0.5:0.95 of our proposed VFD-YOLO model reach 75.0%, 75.6%, 79%, and 54.7%, respectively, which represent improvements of 1.6%, 1.9%, 1.7%, and 2.4% compared with those of the baseline YOLOv13 model. In summary, the model we propose exhibits superior detection performance and can better meet the needs of practical scenarios such as intelligent transportation and urban surveillance. Full article

Cross-organization cyber defense must reconcile collaborative learning with privacy and adversarial robustness, yet standard federated learning ships full gradient tensors, leaking sensitive posture and inviting Byzantine manipulation. We present FedMARL-LTI, a federated multi-agent reinforcement learning framework whose architecture answers both pressures with a single decision: each organization’s threat intelligence is shared only as a differentially private 768-dimensional semantic embedding, never as raw data. In the evaluated system, a Weight-DP-protected model-weight delta is also exchanged through the federated aggregator (the semantic abstraction embedding is a parallel channel); the privacy guarantee below is stated for the semantic abstraction channel, and an embeddings-only architecture—which the guarantee enables—is the design this points toward. The contribution is fourfold. (1) Semantic Abstraction (SA) channel: per organization, each round, the local gradient is summarized by an LLM, projected to a 768-dim embedding, L2-clipped, and Gaussian-noised before any numeric quantity leaves the host. The bottleneck reduces the aggregate noise magnitude—the expected L2 norm of the DP noise vector—from $O\left(\sqrt{d_{\mathrm{model}}}\right)$ to $O\left(\sqrt{m}\right)$ with $m=768\ll d_{\mathrm{model}}\approx 3\times {10}^5$. (2) Formal privacy analysis: the SA + DP cascade satisfies $(\epsilon ,\delta )$-DP and bounds per-round mutual information leakage by $\min \{T_{\mathrm{tok}}{\log }_{2}V,m/2{\log }_2(1+C^2/\left(m{\sigma }^2\right)\left)\right\}$, with Rényi composition over $T$ federation rounds. Scope of the guarantee: this bound certifies (i) the semantic-abstraction channel. It does not by itself cover (ii) the weight-aggregation channel, whose Weight-DP protection is analyzed separately, nor (iii) the whole deployed system, which is the composition of the two. We therefore state the ≈1.4-bit/MI bound as a per-round guarantee on information leaving the organization through the SA channel not over every byte the system emits; an embeddings-only configuration—which this bound enables—closes the gap to a whole-system guarantee. (3) Byzantine-resilient ClippedClustering aggregator combining L2 clipping with cosine-similarity clustering. (4) Hierarchical MARL policy with threat-profile-aware LLM-IRR reward shaping, wired end-to-end and disclosed honestly (the evaluated system uses a deterministic Johnson–Lindenstrauss projection in place of the LLM call for reproducibility; the architecture is thus LLM-compatible rather than dependent on a specific model, and a full LLM deployment is the planned extension). We evaluate on CybORG CAGE-4 with $n=5$ organizations, 30 federation rounds × 5 episodes × 100 steps per round. Releasing the SA channel in parallel shows no statistically detectable reward cost at N = 5 vs. the no-privacy baseline; this is measured at reward-shaping coefficient β = 0, so it establishes that the private semantic release does not disturb weight-channel training rather than that semantic sharing improves defense: SA-only $\Delta$reward = $+4.58$ ($t=+1.37$, NS), dual SA + Weight-DP $\Delta$reward = $+4.31$ ($t=+1.30$, NS), all $N=5$ seeds, all $\left|t\right|<1.4$. A controlled signal/noise probe confirms a $19.58\times$ improvement of SA over Weight-DP at a fixed DP budget—matching the predicted $\sqrt{d/m}\approx 19.8$. Under Byzantine sign_flip at 30% ($N=15$), ClippedClustering is directionally strongest ($F_1=0.025$ vs. FedAvg $0.020$, Krum $0.016$) but the edge is not statistically significant (CC vs. Krum $t=+1.59$, $p=0.15$, $d=+0.58$; the earlier $N=5$ “$3.4\times$” gap was small-sample optimism); its Byzantine behavior is on the harsher random_noise attack. Under a corrected implementation, the undefended baselines do not diverge or collapse; the earlier reading (Krum $0.002$, ClippedClustering $0.020$) was a noise-injection artifact and is withdrawn; ClippedClustering is now directionally best on F1 but not significantly, and trails Krum on reward (superseded Cohen’s $d=+3.77$). The cooperative-PPO family (MAPPO, IPPO) outperforms value/actor-critic (QMIX, MADDPG) by $\approx 20$ reward units, $p<0.001$. All host-level $F_1$ values stay below $0.05$ at the 15K-step training horizon used here; the relative claims of the paper (no detectable privacy reward cost, ClippedClustering’s competitive (not decisive) Byzantine behavior on the harsher attacks, cooperative-PPO dominance) are unaffected by this scope. A 200K-step long-horizon replication lifts $F_1$ above the 15K plateau (to $\approx 0.044$, $N=5$)—confirming that horizon, not the privacy/Byzantine machinery, gates absolute accuracy—but a finer 60-checkpoint run shows the climb is volatile and non-monotonic and does not reach deployment-grade, an honest stability-not-compute limitation. FedMARL-LTI is therefore presented as a proof-of-concept for the relative privacy and robustness trade-offs it isolates, not as an operationally deployable cyber defense system. We release all 141 raw run JSON outputs (Phases 1–3, the L4 backend comparison, and the algorithm/aggregator baselines), the figures, and analysis scripts for replication. Full article

Recent vision backbones increasingly rely on sophisticated modules, whereas long-used operators such as residual connections, activations, and normalization layers remain less explored for lightweight redesign. This paper revisits these operators and proposes three operator-level redesigns: Subtractive Residual Connection (SRC), Learnable Gating Response Function (LGRF), and statistics-discrepancy-guided Dynamic Dual Normalization (DDN). SRC changes shallow residual fusion from addition to subtraction to suppress redundant responses and induce attention-like response focusing without an explicit attention branch; LGRF extends fixed gating activations to channel-wise learnable response curves; and DDN generates sample-level LayerNorm (LN)–BatchNorm (BN) fusion weights from input statistics and LN-BN discrepancy. On ImageNet-1K, SRC improves ResNet models without extra parameters or floating-point operations (FLOPs). Applying all three proposed methods to MambaOut-Femto improves Top-1 accuracy by 1.01 percentage points with only 0.06 M additional parameters and 0.01 GFLOPs. Ablations on ImageNet-100 and CIFAR-10/100 support effectiveness and stability, while DeepWeeds validation further supports the practical value of all three operators for weed recognition under complex natural backgrounds. Gradient-weighted Class Activation Mapping (Grad-CAM), feature response maps, learned response curves, and dynamic LN-BN weights support module interpretation. Overall, revisiting long-used fundamental operators remains valuable for efficient visual recognition. Full article

Conventional histopathological diagnostics rely on morphological assessment of stained tissue sections, requiring extensive sample preparation and subjective expert interpretation. Mid-infrared (MIR) imaging offers a complementary approach by providing spatially resolved, label-free access to the intrinsic biochemical composition of tissue without exogenous contrast agents. This work introduces a preprocessing and analysis pipeline for multiphotometric MIR data, applied to formalin-fixed, paraffin-embedded tissue sections from two patients with histopathologically confirmed head and neck squamous cell carcinoma. Combining differential scattering correction, sub-pixel channel registration, and automated tissue segmentation with unsupervised K-Means clustering, the pipeline achieves label-free discrimination of biochemically distinct tissue compartments. K-Means clustering identified four distinct clusters, of which three corresponded to tissue compartments with protein-to-lipid ratios tentatively consistent with epithelial, tumor-associated, and stromal compartments. The resulting cluster maps showed partial spatial correspondence with mIF reference stainings targeting epithelial and stromal markers, supporting the potential of this approach for label-free tissue characterization in digital pathology. Full article

Various wafer defects appear inevitably, due to the highly complex and precise semiconductor fabrication processes. Thus, precise and rapid detection of patterned wafer surface defects is essential to prevent circuit failures and ensure product quality. Accordingly, a novel lightweight detection network termed PWDD-Net is proposed in this work, by introducing several modifications on YOLO11-nano. First, a novel LGE block, incorporating spatial and channel transformation with adaptive gated mechanism, is developed to enhance fine-grained feature extraction and representation. Second, by integrating a self-calibration block, a lightweight SC-C3k2 module is proposed to improve global feature capture while preserving network efficiency. Finally, the Slide loss is employed to distinguish easy and hard instances, thereby mitigating the imbalanced class distribution and improving classification precision. Experimental results show that PWDD-Net achieves a mAP@0.5 of 74.4% and a mAP@0.5:0.95 of 46.5%, yielding remarkable increments of 4.2% and 2.1% over the YOLO11-nano baseline, respectively. In addition, the network maintains a comparable parameter scale to the baseline and performs an inference speed of 78 FPS using an NVIDIA RTX 3080Ti GPU. These results demonstrate the model’s potential for real-time industrial wafer defect inspection applications. Full article

Variable illumination, severe occlusion, and the small scale of pest instances in complex forest environments significantly compromise the accuracy of automated forest pest detection. Existing deep learning-based detection models often face challenges in achieving an optimal trade-off among detection accuracy, model lightweightness, and training stability. To achieve efficient and intelligent monitoring of forest pests, we propose a novel detection framework named DAAL-YOLOv12n by improving the lightweight YOLOv12n baseline with two specifically designed modules for complex forest scenarios. To overcome the limited feature adaptability caused by fixed convolutional branches and conventional residual connections in the original A2C2f block, a DynamicA2C2f dynamic attention module is introduced. This module employs dynamic gating weights and adaptive residual adaptation modulation to achieve directional feature enhancement and improved cross-channel compatibility. Furthermore, to reduce the computational redundancy of the conventional C3/C2f block and alleviate abnormal optimization errors in the exponential moving average (EMA) caused by persistently stored dynamic tensors, an AdaptiveC3K2 lightweight adaptation module is developed. Through shared convolution representations, temporary dynamic weight generation, and pre-residual adaptation strategies, the proposed module achieves model compression, flexible channel adaptation, and EMA-compatible optimization. Extensive experiments are conducted on a self-collected forest pest dataset comprising 16 common pest categories. Experimental results demonstrate that, compared with advanced detectors such as YOLOv12n, DAAL-YOLOv12n improves precision and recall by 10.5% and 11.1%, respectively, with only a marginal increase in model parameters. Furthermore, the proposed approach increases mAP@0.5 and mAP@0.5:0.95 by 10.2% and 18.0%, respectively, attaining an mAP@0.5 of 93.0% alongside a real-time inference throughput of 93 FPS. These results indicate that the proposed method offers a viable solution for deploying accurate and lightweight forest pest detection systems on edge devices, offering technical support for intelligent pest surveillance in the wild. Full article

To address the need for low-power, real-time underwater acoustic signal processing and autonomous target detection on deep-sea unmanned mobile platforms, such as profiling acoustic buoys and underwater gliders, this study developed an embedded Linux-based signal processing system for buoy platforms. Conventional digital signal processing hardware platforms are often constrained by large size, high power consumption, and limited data communication capability. The proposed system adopts a compact, low-power architecture and a multithreaded processing framework based on the AM6254 heterogeneous multicore processor. It acquires four-channel vector-hydrophone signals together with attitude data from an inertial navigation module and performs band-pass filtering, fast Fourier transform (FFT), direction-of-arrival (DOA) estimation, and constant false alarm rate (CFAR) detection for autonomous target detection. The measured typical power consumption was approximately 2.3 W. Anechoic-tank and sea-trial results showed the lowest tested spectral level at which autonomous detection was achieved was 54 dB at 1 kHz, corresponding to an average in-band level of 46 dB. Under sea state 3, the system maintained continuous bearing tracking after target acquisition for a surface target traveling at 7 kn, up to a range of approximately 7 km, and provided unambiguous bearing estimation. These results demonstrate the target-detection capability and practical applicability of the system under representative operating conditions and indicate its potential for marine environmental monitoring and unmanned-platform observation and detection. Full article