A Visual Cortex-Attentive Deep Convolutional Neural Network for Digital Image Design DOI Creative Commons
Lei Zheng

Journal of Computing and Information Technology, Год журнала: 2024, Номер 31(1), С. 21 - 37

Опубликована: Янв. 8, 2024

With the proliferation of advanced visualization techniques in visual communication, enhancing digital image quality remains a persistent challenge. This study presents sophisticated Convolutional Neural Network (CNN) model to optimize processing. The incorporates multi-stage architecture attentive biological pathways. Inter-subnetwork connections enable integrated feature learning, guided by adaptive weighting luminance, color, orientation, and edge maps. Spatial channel attention modules further enrich interplay. When evaluated on LIVE 3D Phase dataset, approach demonstrates marked improvements, with saliency maps closely mirroring human perception. Pearson Correlation Coefficient Histogram Intersection metrics exceed conventional models, at 0.6486 0.7074 respectively. Testing across distortion types reveals strong agreement subjective rankings, confirming model's effectiveness. By combining automated extraction insights from cortex mechanisms, this bio-inspired CNN framework significantly enhances optimization quality. scalable provides foundation for next-generation computer vision machine learning applications.

Язык: Английский

Soft robotic hand with tactile palm-finger coordination DOI Creative Commons
Ningbin Zhang, Jieji Ren, Yunlong Dong

и другие.

Nature Communications, Год журнала: 2025, Номер 16(1)

Опубликована: Март 10, 2025

Soft robotic hands with integrated sensing capabilities hold great potential for interactive operations. Previous work has typically focused on integrating sensors fingers. The palm, as a large and crucial contact region providing mechanical support sensory feedback, remains underexplored due to the currently limited density interaction Here, we develop sensorized hand that integrates high-density tactile dexterous soft fingers, cooperative palm-finger strategies. palm features compact visual-tactile design capture delicate information. fingers are designed fiber-reinforced pneumatic actuators, each two-segment motions multimodal grasping. These enable extensive interactions, offering mutual benefits such improved grasping stability, automatic exquisite surface reconstruction, accurate object classification. We also feedback strategies dynamic tasks, including planar pickup, continuous flaw detection, pose adjustment. Furthermore, our development, augmented by artificial intelligence, shows human-robot collaboration. Our results suggest promise of fusing rich advanced Robotic operations, yet underexplored. authors present soft, enhancing coordination operation perception.

Язык: Английский

Процитировано

2
LDC: Lightweight Dense CNN for Edge Detection DOI Creative Commons
Xavier Soria Poma,

Gonzalo Pomboza-Junez,

Ángel D. Sappa

и другие.

IEEE Access, Год журнала: 2022, Номер 10, С. 68281 - 68290

Опубликована: Янв. 1, 2022

This paper presents a Lightweight Dense Convolutional (LDC) neural network for edge detection. The proposed model is an adaptation of two state-of-the-art approaches, but it requires less than 4% parameters in comparison with these approaches. architecture generates thin maps and reaches the highest score (i.e., ODS) when compared lightweight models (models 1 million parameters), similar performance compare heavy architectures about 35 parameters). Both quantitative qualitative results comparisons models, using different detection datasets, are provided. LDC does not use pre-trained weights straightforward hyper-parameter settings. source code released at https://github.com/xavysp/LDC.

Язык: Английский

Процитировано

30
Comprehensive review of edge and contour detection: from traditional methods to recent advances DOI
Qinyuan Huang, Jiasheng Huang

Neural Computing and Applications, Год журнала: 2025, Номер unknown

Опубликована: Янв. 5, 2025

Язык: Английский

Процитировано

1
Tiny and Efficient Model for the Edge Detection Generalization DOI
Xavier Soria Poma, Yachuan Li, Mohammad Rouhani

и другие.

Опубликована: Окт. 2, 2023

Most high-level computer vision tasks rely on low-level image operations as their initial processes. Operations such edge detection, enhancement, and super-resolution, provide the foundations for higher level analysis. In this work we address detection considering three main objectives: simplicity, efficiency, generalization since current state-of-the-art (SOTA) models are increased in complexity better accuracy. To achieve this, present Tiny Efficient Edge Detector (TEED), a light convolutional neural network with only 58K parameters, less than 0.2% of models. Training BIPED dataset takes 30 minutes, each epoch requiring 5 minutes. Our proposed model is easy to train it quickly converges within very first few epochs, while predicted edge-maps crisp high quality. Additionally, propose new test which comprises samples from popular images used segmentation. The source code available https://github.com/xavysp/TEED.

Язык: Английский

Процитировано

15
Machine learning methods for the industrial robotic systems security DOI

Dmitry Tsapin,

K. V. Pitelinskiy,

S. V. Suvorov

и другие.

Journal of Computer Virology and Hacking Techniques, Год журнала: 2023, Номер 20(3), С. 397 - 414

Опубликована: Сен. 2, 2023

Язык: Английский

Процитировано

14
Curvilinear object segmentation in medical images based on ODoS filter and deep learning network DOI
Yuanyuan Peng, Lin Pan, Pengpeng Luan

и другие.

Applied Intelligence, Год журнала: 2023, Номер 53(20), С. 23470 - 23481

Опубликована: Июль 12, 2023

Язык: Английский

Процитировано

13
A lightweight contour detection network inspired by biology DOI Creative Commons
Chuan Lin, Zhenguang Zhang,

Jiansheng Peng

и другие.

Complex & Intelligent Systems, Год журнала: 2024, Номер 10(3), С. 4275 - 4291

Опубликована: Март 4, 2024

Abstract In recent years, the field of bionics has attracted attention numerous scholars. Some models combined with biological vision have achieved excellent performance in computer and image processing tasks. this paper, we propose a new bio-inspired lightweight contour detection network (BLCDNet) by combining parallel mechanisms bio-visual information convolutional neural networks. The backbone BLCDNet simulates pathways ganglion cell–lateral geniculate nucleus primary visual cortex (V1) area, realizing step-by-step extraction input information, effectively extracting local features detailed images, thus improving overall model. addition, design depth feature module separable convolution residual connection decoding to integrate output network, which further improves We conducted large number experiments on BSDS500 NYUD datasets, experimental results show that proposed paper achieves best compared traditional methods previous biologically inspired methods. still outperforms some VGG-based without pre-training fewer parameters, it is competitive among all them. research also provides idea for combination

Язык: Английский

Процитировано

5
Process mapping and anomaly detection in laser wire directed energy deposition additive manufacturing using in-situ imaging and process-aware machine learning DOI Creative Commons
Anis Assad, Benjamin Bevans,

Willem Potter

и другие.

Materials & Design, Год журнала: 2024, Номер 245, С. 113281 - 113281

Опубликована: Авг. 28, 2024

This work concerns the laser wire directed energy deposition (LW-DED) additive manufacturing process. The objectives were two-fold: (1) process mapping – demarcating states as a function of processing parameters; and (2) monitoring detecting anomalies (instabilities) using data acquired from an in-situ meltpool imaging sensor. LW-DED enables high-throughput, near-net shape manufacturing. Without rigorous parameter control, however, often introduces defects due to stochastic drifts. To enhance scalability reliability, it is essential understand how parameters affect regimes, detect deleterious In this work, single-track experiments conducted over 128 combinations power, scanning velocity, linear mass density. Four observed via high-speed delineated stable, dripping, stubbing, incomplete melting regimes. Physically intuitive features used train simple machine learning models for classifying state into one four approach was benchmarked against computationally intense, black-box deep that directly use as-received images. Using only six morphology intensity signatures, classified with statistical fidelity approaching 90 % (F1-score) compared F1-score 87 models.

Язык: Английский

Процитировано

3
COS-Net: Bio-inspired Color Opponent and Orientation Selectivity Network for Edge Detection DOI
Zhefei Cai,

Yingle Fan,

Tao Fang

и другие.

Digital Signal Processing, Год журнала: 2025, Номер unknown, С. 104994 - 104994

Опубликована: Янв. 1, 2025

Язык: Английский

Процитировано

0
Contrast-Invariant Edge Detection: A Methodological Advance in Medical Image Analysis DOI Creative Commons
Li Dang, Patrick Cheong-Iao Pang, Charlene Lam

и другие.

Applied Sciences, Год журнала: 2025, Номер 15(2), С. 963 - 963

Опубликована: Янв. 19, 2025

Edge detection methods are significant in medical imaging-assisted diagnosis. However, existing based on grayscale gradient computation still need to be optimized practicality, especially terms of actual visual quality and sensitivity image contrast. To optimize the visualization enhance robustness contrast changes, we propose Contrast Invariant Detection (CIED) method. CIED combines Gaussian filtering morphological processing preprocess images. It utilizes three Most Significant Bit (MSB) planes binary images detect extract edge information. Each bit plane is used edges 3 × blocks by proposed algorithm, then information from each fused obtain an image. This method generalized common types Since eliminates complex pixel operations, it faster more efficient. In addition, insensitive changes contrast, making flexible its application. comprehensively evaluate performance CIED, develop a dataset conduct evaluation experiments these The results show that average precision 0.408, recall 0.917, F1-score 0.550. indicate not only practical effects but also robust invariance. comparison with other confirm advantages CIED. study provides novel approach for within

Язык: Английский

Процитировано

0