Search Results (72)

Background/Objectives: Colorectal cancer (CRC) is the second leading cause of cancer-related mortality worldwide, with polyp miss rates of up to 26% reported during colonoscopy and classification accuracy remaining highly operator-dependent. Accurate multi-class polyp subtype classification is clinically critical, as it directly determines treatment decisions: adenomatous polyps require resection, whereas hyperplastic lesions may warrant only surveillance. This study aims to systematically compare five deep learning architectures for five-class gastrointestinal polyp classification and to provide clinically interpretable diagnostic insights through Grad-CAM visualisation. Methods: ResNet50, VGG16, EfficientNet-B3, DenseNet121, and Vision Transformer (ViT-B/16) were evaluated on the Kvasir Dataset V2 (5000 images, five classes) under a unified training and evaluation protocol on common GPU hardware. All models employed ImageNet transfer learning with a redesigned multi-layer classification head. Two optimisation strategies were applied: SGD with cosine annealing for CNN architectures, and AdamW with linear warmup for ViT-B/16. Gradient-weighted Class Activation Mapping (Grad-CAM) was applied to generate spatial attention heatmaps for qualitative clinical interpretation. Results: Under a single 80/10/10 split, ViT-B/16 attained the highest accuracy (97.2%); however, because a single split is sensitive to sampling, the evaluation was strengthened with stratified five-fold cross-validation (mean ± SD). Under cross-validation, EfficientNet-B3 achieved the highest accuracy at 95.90 ± 0.35%, followed closely by ViT-B/16 (95.12 ± 0.72%), then ResNet50 (91.74 ± 0.74%), DenseNet121 (90.32 ± 0.70%), and VGG16 (88.50 ± 1.72%); the small standard deviations indicate that all models, including ViT-B/16, were stable across folds. Pairwise McNemar tests with Holm correction found that every difference was statistically significant (p < 0.05), including the EfficientNet-B3 advantage over ViT-B/16 (p = 0.010). ViT-B/16 thus remained a strong, stable performer that significantly outperformed the three remaining CNNs, while the cross-validated ranking placed the most compact model, EfficientNet-B3, first: a Vision Transformer was highly competitive with, but not superior to, the strongest CNN. A consistent, architecture-agnostic misclassification pattern was identified between dyed-lifted polyps and dyed-resection margins across all five models, consistent with a task-level visual ambiguity that may also reflect overlapping class definitions and annotation factors, with direct clinical implications. Grad-CAM analysis, quantified by attention-entropy and concentration metrics, showed that model attention remained focused on relevant stained tissue regardless of whether predictions were correct, indicating that the dyed-class confusions reflect genuine visual ambiguity rather than a localisation failure. Conclusions: Under cross-validation, EfficientNet-B3 achieved the highest accuracy on the Kvasir V2 five-class task, significantly outperforming all other architectures, with ViT-B/16 being a close and competitive second. The identified confusion between post-procedural chromoendoscopic classes is unlikely to be fully resolved by architectural changes alone and may require higher-resolution imaging or domain expert re-annotation. These findings contribute to the evidence base for explainable deep learning in gastrointestinal endoscopy; external, multi-centre validation remains necessary before clinical adoption. Full article

In renewable-powered distribution systems and microgrids, reliable cable condition monitoring is essential for operational security and early fault detection. Sheath circulating current signals provide valuable information for identifying grounding abnormalities and incipient faults, but their diagnosis is difficult because the signals exhibit strong inter-phase coupling, and fault samples are limited in practice. This study proposes a dual-view Mixup-ResNet framework for fault diagnosis of cable sheath circulating current signals. Specifically, physical-range normalization is employed to retain magnitude-related fault information and inter-phase proportional relationships, while sample-wise z-score normalization is used to emphasize waveform morphology. These two complementary views are concatenated to form a six-channel input for a lightweight one-dimensional residual network, which is trained with Mixup, label smoothing, dropout, and cosine annealing. On an ATP-EMTP-generated eight-class dataset, the proposed method achieves an average accuracy of 91.50%, a weighted F1-score of 91.69%, and a macro F1-score of 91.69% under a unified 5 × 5 repeated stratified cross-validation protocol. Additional tests under 12% relative Gaussian noise show that the method maintains competitive Gaussian-noise tolerance within the tested condition, although broader field disturbances remain to be further validated. These findings suggest that the proposed method has potential for small-sample fault diagnosis of cable sheath circulating current signals and provides a preliminary basis for intelligent cable condition monitoring. Full article

Accurate ultra-short-term wind power forecasting is of great significance for grid integration scheduling and the secure operation of power systems. However, due to meteorological disturbances and turbine operating states, wind power series generally exhibit non-stationary, multi-scale fluctuations and strong nonlinearity. To improve forecasting accuracy, this paper proposes an ultra-short-term wind power forecasting model based on a two-stage signal decomposition and a hybrid architecture combining iTransformer, LSTM, and KAN. First, a cascaded decomposition module is constructed using the wavelet transform (WT) and ICEEMDAN to attenuate the non-stationarity of the original power series and to extract multi-scale features. An iTransformer branch is then employed to model global dependencies among multiple variables, while an LSTM branch captures temporal dynamics in the historical power series. Subsequently, a cross-attention mechanism is introduced to achieve cross-branch feature fusion, and a KAN output layer is adopted to enhance the model’s representation of the wind speed–power nonlinear mapping. A particle swarm optimization (PSO) algorithm, combined with a cosine annealing strategy, is used to optimize key hyperparameters and improve training stability. Experimental results using SCADA data from a 150 MW wind farm in southern Hunan Province show that the proposed model achieves an MAE of 9.8327 MW, an RMSE of 13.1872 MW, an SMAPE of 18.8474%, and an $R^2$ of 0.7798. These values correspond to the fixed main comparison protocol used for baseline evaluation, while the ablation study reports multi-seed mean and standard deviation results to assess module-level robustness. Compared with LSTM and WT-ICEEMDAN-CNN-LSTM, the proposed model achieves clear improvements in forecasting accuracy and fitting capability. Additional cross-wind-farm validation on a second wind farm shows that WT-ICEEMDAN-iTransformer-LSTM-KAN-PSO (hereafter referred to as ILKP) maintains the best overall performance, achieving an MAE of 27.2193 MW, an RMSE of 36.1862 MW, an SMAPE of 27.8429%, and an $R^2$ of 0.5189, demonstrating transferability and robustness under different operating conditions. Full article

This paper tackles the problem of performance limitations on channel estimation accuracy in near-field massive MIMO systems resulting from the sparse reconstruction method with fixed spatial grid dictionaries. It presents a deep unfolded network with a learnable sparse dictionary (LSD-UnfoldNet) to address the grid mismatch, thereby obtaining high accuracy and low-complexity near-field channel estimation as the sparse dictionary and channel reconstruction parameters are jointly optimized using end-to-end learning. The iterative shrinking threshold algorithm is unfolded into an L-layer neural network with L learnable sparse dictionaries and linear transformation matrices in each layer of the network. Besides, to improve the correlation characteristics of the sensing matrix, an orthogonal regularization term is introduced. During training of the neural network, the Adam optimizer and cosine annealing learning rate scheduling are applied to jointly minimize the normalized mean square error and total loss of learned dictionary correlations. Experimental results demonstrate that the proposed method, at a representative signal-to-noise ratio of 20 dB, achieves normalized mean square errors of −24.5 dB. The root mean square errors for angle and distance estimation were 0.50° and 0.18 m, respectively, achieving superior performance relative to heuristic algorithms and deep learning-based algorithms. Beyond its strong robustness to different grid granularities and spatial distances, the proposed method also achieves competitive pilot overhead compared with other approaches. Full article

This study proposes TPA-ConvNeXt, a ConvNeXt-Small-based deep learning architecture for retinal image classification using a phase-based trigonometric attention mechanism. The proposed Trigonometric Phase Attention (TPA) block recalibrates feature maps through learnable phase and amplitude modulation derived from channel-wise and spatial context. In addition, a stable fine-tuning strategy combining layer-wise learning rate decay (LLRD), linear warmup, and cosine annealing is employed to adapt pretrained backbones to medical image data. The method was evaluated on both fundus and optical coherence tomography (OCT) datasets. On the HYAMD fundus dataset, 5-fold cross-validation yielded a mean Macro F1 of 0.8940 ± 0.0403 and an ROC-AUC of 0.9449. External fundus validation further demonstrated cross-dataset robustness, achieving 0.9312 Macro F1 and 0.9803 ROC-AUC when trained on HYAMD and tested on AMDNet23. In OCT experiments, the model achieved Macro F1 scores of 0.9790 on MAK1_OCT, 0.9387 on OCTDL, and 0.9989 on the cleaned OCT2017 benchmark test set after MD5 duplicate removal. Ablation results showed that warmup contributed most strongly to stable optimization, while the proposed TPA block provided a smaller but consistent performance gain. Additional statistical analysis across the five HYAMD folds showed that TPA-ConvNeXt achieved comparable performance to the ConvNeXt-Small baseline, with a small numerical accuracy difference that did not reach statistical significance. Grad-CAM visualizations indicated that the model focused on clinically relevant retinal regions, and complexity analysis showed that the full model required 51.82 M parameters and 17.41 GFLOPs, adding only 2.36 M parameters and 0.02 GFLOPs over the ConvNeXt-Small baseline. These findings suggest that TPA-ConvNeXt provides a robust and generalizable framework for retinal image classification across both fundus and OCT modalities. Full article

UAV path planning in complex static 3D environments involves multiple conflicting objectives and intricate constraints. However, when applied to highly constrained path planning tasks, MOWOA often suffers from a low proportion of feasible solutions, convergence instability, single-leader search bias, and an uneven distribution of Pareto solutions. To address these issues, this study formulates the UAV path planning problem as a multi-objective optimization problem that simultaneously considers path length, threat cost, smoothness cost, and altitude cost, and proposes an improved multi-leader multi-objective whale optimization algorithm (IML-MOWOA). The proposed IML-MOWOA progressively improves three key stages of the optimization process: initial population construction, search guidance, and external archive maintenance. Specifically, an adaptive opposition-based learning initialization strategy is first introduced to improve the feasibility and spatial coverage of initial paths. Based on the resulting non-dominated solution set, a grid-based external archive update strategy is then used to regulate solution density and provide representative candidate leaders from sparse Pareto regions. Subsequently, a multi-leader dynamic weighted search mechanism with Softmax-based cosine annealing integrates these leaders into the WOA update process, thereby enhancing multi-directional path exploration and alleviating premature convergence. Comparative experiments conducted in three static 3D environments of varying complexity demonstrate that the proposed method achieves more robust convergence, better Pareto-front distribution, and more balanced task-level path quality than the benchmark algorithms. In the most challenging scenario, IML-MOWOA achieves the highest number of feasible paths, reduces the mean IGD by 25.04%, and decreases the mean path length, threat cost, smoothness cost, and altitude cost by 1.65%, 28.45%, 53.23%, and 29.88%, respectively, compared with the best-performing competing algorithm for each metric. These results indicate that the proposed algorithm is effective and robust for constrained multi-objective UAV path planning in complex static 3D environments. Full article

Accurate, automated delineation of adult diffuse gliomas from multi-parametric magnetic resonance imaging (mpMRI) is central to quantitative neuro-oncology. Volumetric 3D networks dominate the BraTS leaderboard but require expensive GPUs, long training cycles, and provide diminishing returns relative to their compute budget. Slice-wise 2D models, by contrast, discard inter-slice context that is informative for thin tumor rims and small enhancing foci. We introduce GDNet, a 2.5D multimodal MRI segmentation framework for adult glioma evaluated on the BraTS 2024 cohort. GDNet consumes a stack of three adjacent axial slices from the four standard BraTS modalities (T1, T1ce, T2, FLAIR) as a 12-channel input to a compact U-shaped encoder–decoder with Group Normalization and predicts whole tumor (WT), tumor core (TC), and enhancing tumor (ET) masks for the central slice. The training pipeline pairs the 2.5D backbone with: (i) Exponential Moving Average (EMA) of model weights with decay 0.999, (ii) mixed tumor-aware slice sampling (p_tumor = 0.50), (iii) a compound Cross-Entropy + Soft-Dice loss, and (iv) AdamW with warm-up plus cosine annealing under Automatic Mixed Precision. We performed a systematic, step-by-step ablation covering a 2D baseline, EMA + mixed sampling, tumor-centered crop fine-tuning, a GDNet-inspired architectural integration, a region-aware loss, 3-slice and 5-slice 2.5D inputs, and connected-component post-processing, and we report multi-seed results to quantify reproducibility. On the held-out BraTS 2024 test partition, the final 3-slice 2.5D GDNet achieved positive-only Dice scores of 0.791 ± 0.000 (WT), 0.736 ± 0.003 (TC), 0.654 ± 0.004 (ET), and a mean foreground positive-only Dice of 0.820 ± 0.000 across seeds; the all-slice mean foreground Dice exceeded 0.927 ± 0.000. Validation positive-only scores were 0.805 ± 0.002 (WT), 0.757 ± 0.004 (TC), 0.683 ± 0.009 (ET). The inter-seed standard deviation was small for every region (≤0.01 Dice points), indicating low inter-seed variance across the two seeds evaluated; with only two seeds, we regard this as preliminary evidence of training stability rather than a strong reproducibility claim. The ablation isolated EMA + mixed tumor sampling and the 2.5D context window as the dominant sources of improvement; notably, a GDNet-style architectural integration with a region-aware loss did not outperform the simpler 2.5D U-Net on positive-only WT/TC/ET, and light post-processing improved only all-slice Dice. A failure-mode audit found that the residual catastrophic predictions are concentrated on a small minority of diffuse, infiltrative tumors with mass effect. Conclusions: Carefully engineered training strategies, tumor-aware sampling, EMA stabilization, and a modest 2.5D context window recover a substantial fraction of the accuracy of much heavier 3D networks at a fraction of the compute, are reproducible across seeds, and outperform a heavier GDNet-inspired architectural variant on the same data. GDNet is therefore a practical and, pending external validation, potentially clinically deployable framework for multimodal glioma segmentation on workstation-class GPU hardware. Full article

It is well-known that activation functions are crucial in determining spectral expressiveness, training dynamics, and reconstruction accuracy in implicit neural representations (INRs), which employ coordinate-based multilayer perceptrons to represent continuous signals. Despite showing excellent performance, sinusoidal activations, for example SIREN, are limited in their adaptability to diverse signal types due to their fixed harmonic structure. In this paper, we propose two novel periodic activation functions for INRs. (1) Harmonic generalizes sinusoidal activations by combining the fundamental frequency with learned second and third harmonics through per-neuron trainable amplitude coefficients, resulting in a richer spectral basis within the SIREN initialization framework. (2) PM-FINER (Phase-Modulated FINER) extends the variable-periodic FINER activation by embedding frequency modulation synthesis directly into the instantaneous phase, enabling data-driven phase distortion via a learnable modulation index and carrier ratio. We conducted comprehensive experiments spanning nine architectural configurations (including SIREN, WIRE, FINER, Gaussian, Harmonic, PM-FINER, and an additional direct comparison against the Subtractive Modulative Network (SMN)), using six natural images, three learning rate schedulers, and three random seeds, totaling 486 main training runs (534 runs total including an ${\omega }_0$ sensitivity sweep). Our evaluation combined peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and rigorous statistical analysis, such as paired t-tests, Wilcoxon signed-rank tests, Cohen’s d effect sizes, and Friedman rank tests. Under cosine annealing, Harmonic achieves a mean PSNR gain of 6.08 dB over SIREN and 2.57 dB over FINER (both $p<0.001$, Cohen’s $d>3.7$), while PM-FINER ranks statistically on par with Harmonic (mean difference $0.17$ dB, $p=0.36$), outperforming all of the other baselines. Compared with SMN, Harmonic outperforms it by $+7.94$ dB under cosine annealing (Bonferroni-adjusted $p<{10}^{-5}$, Cohen’s $d=12.3$), winning on all six images. Additionally, the Friedman ranking across the six images confirmed Harmonic (with mean rank $=1.33$) and PM-FINER (with mean rank $=1.67$), being the top two methods under cosine annealing. Our results establish interpretable multi-harmonic and phase-modulated activations as real alternatives to the existing INR activation functions. Full article

High dynamic range (HDR) image reconstruction from a single low dynamic range (LDR) input remains an important problem for computational photography, particularly when practical deployment on consumer-grade hardware is considered. With the increasing availability of hardware supporting HDR, public demand for capturing and viewing HDR images has grown significantly. Recent research has explored deep learning-based approaches to reconstruct HDR images from low dynamic range (LDR) inputs by extracting regional pixel features or leveraging the camera response function (CRF) for model training. Many of these approaches employ Convolutional Neural Network (CNN) architectures and utilize skip connections to preserve learned information. Nevertheless, the configuration-level effects of data augmentation in HDR reconstruction remain insufficiently discussed. Existing CNN-based approaches, such as HDRCNN, HDRUNet, and ExpandNet, have demonstrated promising reconstruction ability, but they may involve a heavy backbone architecture, a long training time, or a limited discussion of how preprocessing configurations affect reconstruction performance. This study presents an engineering-oriented HDR reconstruction framework derived from HDRCNN, focusing on practical efficiency, structural fidelity, and training feasibility. The proposed framework introduces three modifications: (1) a configuration-level comparison of composite data augmentation settings, including unsharp masking, denoising, Gaussian blur, and brightness–contrast adjustment; (2) the replacement of the original VGG16 backbone with a ResNet50-based encoder enhanced with attention blocks and squeeze-and-excitation (SE) blocks for improved multi-scale feature extraction and channel-wise recalibration; and (3) the integration of mixed-precision training with cosine annealing learning-rate scheduling to reduce computational cost. Experimental results on the SI-HDR dataset show that the best composite augmentation configuration improves PSNR from 19.05 dB to 22.10 dB and SSIM from 0.6444 to 0.7714 without increasing the training time. Compared with the original VGG16-based HDRCNN setting, the ResNet50-based model reduces training time while improving SSIM from 0.2705 to 0.8512. Under the adopted comparison protocol, the proposed model achieves the shortest training time and slightly higher PSNR than HDRUNet, while HDRUNet retains a higher SSIM. This indicates a trade-off among pixel-wise fidelity, structural similarity, and computational efficiency. The current evaluation is limited by a small test setting, composite rather than operation-level augmentation analysis, and the use of PSNR and SSIM only; therefore, future work should include full benchmark evaluation, additional perceptual/HDR-specific metrics, and controlled component-level ablation studies. Full article

Electrocardiogram (ECG) analysis is a cornerstone non-invasive diagnostic technique for detecting cardiac arrhythmias, which remain a leading cause of mortality worldwide. While recent advances in deep learning have significantly improved automated arrhythmia classification, the current literature lacks systematic, fair comparisons of fundamental neural architectures under unified experimental conditions, and very few studies provide model interpretability. This study addresses these gaps by first providing a rigorous comparative analysis of three representative architectures—Artificial Neural Network (ANN), Convolutional Neural Network (CNN), and Residual Network (ResNet)—on the MIT-BIH Arrhythmia Database under identical preprocessing, training, and evaluation protocols. We then propose an efficient Fine-Tuned CNN (FT-CNN) optimized for ECG signal characteristics through adaptive kernel sizing for P-QRS-T morphological extraction, multi-faceted regularization including L2, dropout, and batch normalization, cosine annealing learning rate, and a custom loss function combining weighted categorical cross-entropy with focal loss with gamma equal to 2.0 to address severe class imbalance. The FT-CNN achieves an accuracy of 98.51%, outperforming fourteen benchmark models, including standard CNN with an accuracy of 97.20%, ResNet with 96.88%, LSTM with 96.50%, GRU with 96.30%, and traditional classifiers. Comprehensive ablation studies confirm an improvement of 6.17% over the baseline. Class-wise analysis reveals excellent performance for normal beats with an F1-score of 0.99, ventricular ectopic beats with 0.95, and unknown beats with 0.98, while supraventricular ectopic beats with an F1-score of 0.79 and fusion beats with 0.70 remain challenging. Unlike most prior works, we integrate Grad-CAM and Integrated Gradients for explainability, quantitatively evaluating attribution faithfulness, sanity checks, and noise robustness. Full article

Traditional woven fabric identification heavily depends on manual experience and subjective judgment, which limits its applicability in modern intelligent textile manufacturing. To address this issue, this paper proposes improved woven fabric recognition approaches based on the ResNet50 and ConvNeXt architectures. For the classical ResNet50, the squeeze-and-excitation network (SENet) and the convolutional block attention module (CBAM) attention mechanisms are integrated separately to boost feature representation, and adopt the Adam optimizer to accelerate convergence. For ConvNeXt, we optimize the network stacking blocks and design a warmup + cosine annealing learning rate scheduling strategy. With the transfer learning strategy, classification experiments are conducted on a self-constructed fabric dataset. The improved ResNet50 model achieves a recognition accuracy of 90.15%, while the optimized ConvNeXt model reaches 90.01%. The models outperform their baseline counterparts, demonstrating the effectiveness of the proposed improvements. This study provides a feasible reference for the research of automatic woven fabric classification and related intelligent textile inspection. Full article

Fish disease and species identification is critical for intelligent aquaculture, directly influencing productivity, sustainability, and economic viability. However, existing approaches largely treat species identification and pathological classification as independent tasks, limiting their ability to capture interdependent features under complex real-world conditions such as occlusion, low contrast, dynamic backgrounds, and high inter-class similarity. Moreover, challenges including class imbalance, cross-species variability, and fine-grained feature discrimination remain insufficiently addressed. To overcome these limitations, this paper proposes a hybrid ConvNeXt–BiLSTM–multi-head self-attention (MHSA) framework for joint fish species and disease classification, where a ConvNeXt-Small backbone extracts hierarchical spatial features that are transformed into a structured sequence and processed by a bidirectional LSTM to capture contextual dependencies, followed by an MHSA module for adaptive feature refinement. An auxiliary species classification branch is incorporated to provide multi-task regularization without additional inference costs. The training pipeline integrates CLAHE-based image enhancement, square-root inverse-frequency focal loss, targeted minority oversampling, and a two-stage progressive learning strategy with differential-rate cosine annealing, complemented by five-view test-time augmentation. For practical deployment, a YOLOv8s detector is employed for fish localization prior to classification. The experimental results demonstrate that the proposed model achieves superior performance, attaining overall top-1 classification accuracy of 94.33%, precision of 97.1%, recall of 90.9%, 96.1% mAP50, and an F1-score of 0.9264, while achieving a macro AUC of 0.994 and maintaining high computational efficiency (213.3 FPS), demonstrating a robust and efficient solution for real-time fish disease screening. Full article

This study presents HGRN2-based Flexible Dynamic Encoder Personal VAD (FDE-HGRN2), a recurrent framework for personal voice activity detection (PVAD). Building on the original LSTM-based FDE-RNN backbone, we replace all recurrent modules with the recently introduced HGRN2 gated linear RNN and adopt a cosine-annealing learning rate schedule to improve both detection accuracy and efficiency. HGRN2 uses gated linear recurrence with non-parametric state expansion, enlarging the recurrent state without increasing the number of trainable parameters and enabling more expressive long-range temporal modeling than conventional LSTMs. We evaluate FDE-HGRN2 on a LibriSpeech-derived PVAD benchmark, where multi-speaker mixtures are constructed by concatenating one to three speakers per utterance and randomly designating a target speaker, following established PVAD data construction practices to ensure direct comparability with prior work. The system uses 40-dimensional Mel-filterbank features as acoustic inputs and conditions the detector on 256-dimensional d-vector embeddings extracted from a pretrained speaker verification network. Experimental results show that FDE-HGRN2 consistently outperforms the original FDE-RNN baseline and several state-of-the-art PVAD models in terms of mean Average Precision and frame-level accuracy, while reducing the parameter count of the recurrent backbone by roughly 15% and yielding substantially smaller models than many competing systems. These findings indicate that HGRN2 provides a more temporally expressive and parameter-efficient alternative to LSTM for PVAD, offering a favorable accuracy–efficiency trade-off for real-world, deployment-oriented personalized speech interfaces. Full article

This paper proposes a Black Kite Algorithm (BKA)-based hyperparameter optimization method for Artificial Neural Network (ANN) training, mitigating local minimum issues associated with conventional training techniques. The resulting BKA-ANN model is then employed to estimate PID controller parameters for DC motor speed regulation. A large-scale dataset of 100,000 samples was generated via MATLAB simulation, with reference speed and load torque stochastically varied, and optimal PID parameters determined by minimizing the ITAE criterion for each operating condition. The optimized controller was evaluated under various operating conditions including transient response, frequency domain analysis (phase margin and bandwidth), parametric robustness, and load disturbance suppression, along with control effort and energy consumption assessments. The proposed BKA-ANN approach was benchmarked against nine algorithms: hybrid atom search optimization-simulated annealing (hASO-SA), harris hawks optimization (HHO), Henry gas solubility optimization with opposition-based learning (OBL/HGSO), atom search optimization (ASO), henry gas solubility op-timization (HGSO), stochastic fractal search(SFS), grey wolf optimization (GWO), sine–cosine algorithm (SCA), and Standard ANN. Simulation results indicate that BKA-ANN achieves stable performance across all tested scenarios, with minimal oscillation and competitive settling time compared to the evaluated algorithms. Full article

The proliferation of next-generation Internet infrastructures and the Internet of Things (IoT) has exponentially increased network traffic complexity. While deep learning (DL)-based intrusion detection systems (IDSs) show immense potential, they persistently suffer from challenges including high computational overhead, vanishing gradients in deep architectures, and acute sensitivity to noise. Consequently, these issues impede their real-time deployment in resource-constrained edge computing environments. To overcome these limitations, we propose a novel, lightweight, and robust intrusion detection framework based on deep neural networks (DNNs). Initially, we employ a Robust Scaler-based statistical preprocessing strategy to supersede traditional Z-score standardization, effectively mitigating the adverse impacts of outliers and burst traffic noise. Subsequently, we design an advanced architecture that integrates self-normalizing residual blocks with a channel attention mechanism. Leveraging compressed hidden layers alongside the Scaled Exponential Linear Unit (SELU) activation function, this architecture not only mitigates the vanishing gradient problem but also amplifies critical traffic features. Concurrently, it achieves a substantial reduction in both parameter count and inference latency. Furthermore, we introduce a cosine annealing strategy to dynamically adjust the learning rate during training, thereby facilitating the model’s escape from local optima and accelerating convergence. Extensive experiments on standard benchmark datasets demonstrate that our proposed framework achieves superior detection accuracy while maintaining exceptional computational efficiency compared to state-of-the-art baselines. Full article