Browse Articles
Discover research articles across all indexed journals
Effects of preoperative anti-PD-1 therapy on intraoperative and postoperative analgesia in patients undergoing radical esophagectomy: a retrospective cohort study
Analyzing Orbital Angular Momentum (OAM) mode via localized beam sampling
Response of urban lake water quality to monthly hydro-meteorological drivers at the catchment scale
High expression of STAT3 and PD-L1 predicts poor prognosis for laryngeal squamous cell carcinoma
Rice straw and single-use plastic waste co-pyrolyzed plastic-char composite for the partial replacement of cement in concrete: a life cycle perspective
Electrocardiographic signs to differentiate between chronic thromboembolic pulmonary hypertension and intermediate-high risk acute pulmonary embolism
Prediction of infiltration degree of ground-glass nodules using a fusion of CT radiomics and deep learning
Improved dung beetle optimized MPPT with novel FOTPIDn(1 + PD) control in grid connected PV for optimal power sharing
A frequency analysis of filterbank initialisation and noise augmentation for LEAF
Abstract Differentiable frontends, such as the LEArnable Frontend (LEAF), have drawn increasing interest from the computer audition (CA) community combining the rigour of traditional signal processing techniques with the flexibility and potential of end-to-end deep learning approaches. Concretely, they promise the ability to automatically learn task-specific features, resulting in both higher performance and better interpretability of CA applications. With the adaptability of LEAF’s parameters being questioned in recent literature, we further dig into the reasons why LEAF does not adjust its parameters. We thus perform a detailed analysis investigating the effects of filterbank initialisation for LEAF in a wide, previously unmatched range of computer audition tasks, namely speech recognition, speech emotion recognition, acoustic scene classification, and bird activity detection. In line with literature, we report that performance stays constantly high irrespective of filterbank initialisation, so long as it covers the entire frequency spectrum, in which case adaptation is minimal. Crucially, however, a filterbank initialised with all frequency bands equally does change its centre frequencies and bandwidths, yet remains with a lower performance. This effect is seemingly independent of how information is spread across frequencies, as we confirm in an additional set of experiments with controlled frequency distributions. This points towards the critical role of initialisation and the inductive bias of LEAF and manifests concerns about the adaptability and interpretability of LEAF across many settings. The code for our experiments is publicly available under https://github.com/millinma/LEAFFrequencyAnalysis .
Design of electronically tunable fractional-order elements based on distributed MOS transistor structures
Optimization of inter-basin water transfer projects: from macroeconomic impacts to technical hydraulic design
Investigating the protective effect of zinc oxide–gallic acid nanoparticles against depression like behavior and memory impairment in animals treated with D-galactose
A multi-cognitive PCB defect detection model integrating Mamba
Abstract Printed Circuit Boards (PCBs) pose significant challenges for defect detection due to their complex textures, small defect targets, and subtle inter-class similarities. Traditional inspection methods are limited in robustness, while many deep learning-based detectors struggle with insufficient tiny-target feature extraction, low feature utilization, and high model complexity. To address these limitations, this study introduces PCB-MMF, a multi-cognitive hybrid framework integrating the Mamba state space model. The proposed MM-NET backbone combines CNN-based local feature extraction with Mamba-based global modeling, augmented by a Three-Stage Multi-Receptive Module (TSMR) to fuse global and multi-scale features while mitigating redundancy. A Multi-Cognitive Visual Augmentation Module (MC-VAM) enhances attention to critical regions and preserves shallow features through residual connections, while a Lightweight Group-Shared Detection Head (LGSD) applies parameter sharing to reduce computational cost without compromising accuracy. Experimental results on HRIPCB, DeepPCB, and DsPCBSD+ datasets demonstrate that PCB-MMF achieves mAP50 scores of 93.43%, 98.68%, and 85.39%, respectively. Furthermore, additional generalization experiments on the NEU-DET dataset achieve an mAP50 of 76.69%, confirming the robust performance of PCB-MMF across different industrial scenarios. Compared with the YOLO11 model, PCB-MMF reduces the number of parameters by 8.9% (from 2.58 to 2.35M) and the computational load (FLOPs) by 12.70%. Relative to the Mamba-YOLO model, PCB-MMF reduces the number of parameters by 58.48% and the computational load (FLOPs) by 55.28%. These findings confirm that PCB-MMF offers a favorable balance between accuracy and efficiency, providing a promising solution for lightweight, high-precision PCB defect detection in industrial applications.