Biomimetic Vision for Zoom Object Detection Based on Improved Vertical Grid Number YOLO Algorithm DOI Creative Commons
Xinyi Shen, Guolong Shi, Huan Ren

et al.

Frontiers in Bioengineering and Biotechnology, Journal Year: 2022, Volume and Issue: 10

Published: May 20, 2022

With the development of bionic computer vision for images processing, researchers have easily obtained high-resolution zoom sensing images. The drones equipped with high-definition cameras has greatly increased sample size and image segmentation target detection are important links during process information. As biomimetic remote usually prone to blur distortion in imaging, transmission processing stages, this paper improves vertical grid number YOLO algorithm. Firstly, light shade a were abstracted, grey-level cooccurrence matrix extracted feature parameters quantitatively describe texture characteristics image. Simple Linear Iterative Clustering (SLIC) superpixel method was used achieve light/dark scenes, saliency area obtained. Secondly, model segmenting dark scenes established made dataset meet recognition standard. Due refraction passing through lens other factors, difference contour boundary value between pixel background would make it difficult detect target, pixels main part separated be sharper edge detection. Thirdly, algorithm an improved proposed real time on processed array. adjusted aspect ratio modified grids network structure by using 20 convolutional layers five maximum aggregation layers, which more accurately adapted "short coarse" identified object information density. Finally, comparison mainstream algorithms different environments, test results aid showed that high spatial resolution images, higher accuracy than had real-time performance accuracy.

Language: Английский

Exploiting dynamic spatio-temporal graph convolutional neural networks for citywide traffic flows prediction DOI
Ahmad Ali,

Yanmin Zhu,

Muhammad Zakarya

et al.

Neural Networks, Journal Year: 2021, Volume and Issue: 145, P. 233 - 247

Published: Oct. 29, 2021

Language: Английский

Citations

257
Genetic Algorithm-Based Trajectory Optimization for Digital Twin Robots DOI Creative Commons
Xin Liu, Du Jiang, Bo Tao

et al.

Frontiers in Bioengineering and Biotechnology, Journal Year: 2022, Volume and Issue: 9

Published: Jan. 10, 2022

Mobile robots have an important role in material handling manufacturing and can be used for a variety of automated tasks. The accuracy the robot’s moving trajectory has become key issue affecting its work efficiency. This paper presents method optimizing mobile robot based on digital twin robot. is created by Unity, trained virtual environment applied to physical space. simulation training provides schemes actual movement Based data returned robot, preset dynamically adjusted, which turn enables correction contribution this use genetic algorithms path planning robots, optimization reducing error through interaction real data. It map learning domain

Language: Английский

Citations

113
Multi-Scale Feature Fusion Convolutional Neural Network for Indoor Small Target Detection DOI Creative Commons
Li Huang, Cheng Chen, Juntong Yun

et al.

Frontiers in Neurorobotics, Journal Year: 2022, Volume and Issue: 16

Published: May 19, 2022

The development of object detection technology makes it possible for robots to interact with people and the environment, but changeable application scenarios make accuracy small medium objects in practical low. In this paper, based on multi-scale feature fusion indoor target method, using device collect different images angle, light, shade conditions, use image enhancement set up amplify a date set, SSD algorithm layer its adjacent features fusion. Faster R-CNN, YOLOv5, SSD, models were trained an scene data transfer learning. experimental results show that can improve all kinds objects, especially relatively scale. addition, although speed improved decreases, is faster than which better achieves balance between speed.

Language: Английский

Citations

97
Self-Tuning Control of Manipulator Positioning Based on Fuzzy PID and PSO Algorithm DOI Creative Commons
Ying Liu, Du Jiang, Juntong Yun

et al.

Frontiers in Bioengineering and Biotechnology, Journal Year: 2022, Volume and Issue: 9

Published: Feb. 11, 2022

With the manipulator performs fixed-point tasks, it becomes adversely affected by external disturbances, parameter variations, and random noise. Therefore, is essential to improve robust accuracy of controller. In this article, a self-tuning particle swarm optimization (PSO) fuzzy PID positioning controller designed based on control. The quantization scaling factors in algorithm are optimized PSO order achieve high robustness manipulator. First all, mathematical model developed, designed. A PD control strategy with compensation for gravity used system. Then, parameters dynamically minute-tuned 1. Through closed-loop loop adjust magnitude factors–proportionality online. Correction values outputted modified 2. factor–proportion factor online achieved find optimal Finally, performance improved verified simulation environment. results show that transient response speed, tracking accuracy, follower characteristics system significantly improved.

Language: Английский

Citations

95
Low-Illumination Image Enhancement Algorithm Based on Improved Multi-Scale Retinex and ABC Algorithm Optimization DOI Creative Commons
Ying Sun, Zichen Zhao, Du Jiang

et al.

Frontiers in Bioengineering and Biotechnology, Journal Year: 2022, Volume and Issue: 10

Published: April 11, 2022

In order to solve the problems of poor image quality, loss detail information and excessive brightness enhancement during in low light environment, we propose a low-light algorithm based on improved multi-scale Retinex Artificial Bee Colony (ABC) optimization this paper. First all, makes two copies original image, afterwards, irradiation component is obtained by used structure extraction from texture via relative total variation for first combines it with obtain reflection which are simultaneously enhanced using histogram equalization, bilateral gamma function correction filtering. next part, second equalization edge-preserving Weighted Guided Image Filtering (WGIF). Finally, weight-optimized fusion performed ABC algorithm. The mean values Information Entropy (IE), Average Gradient (AG) Standard Deviation (SD) images respectively 7.7878, 7.5560 67.0154, improvement compared 2.4916, 5.8599 52.7553. results experiment show that proposed paper improves problem process, enhances sharpness, highlights details, restores color also reduces noise good edge preservation enables better visual perception image.

Language: Английский

Citations

83
Real-Time Target Detection Method Based on Lightweight Convolutional Neural Network DOI Creative Commons
Juntong Yun, Du Jiang, Ying Liu

et al.

Frontiers in Bioengineering and Biotechnology, Journal Year: 2022, Volume and Issue: 10

Published: Aug. 16, 2022

The continuous development of deep learning improves target detection technology day by day. current research focuses on improving the accuracy technology, resulting in model being too large. number parameters and speed are very important for practical application embedded systems. This article proposed a real-time method based lightweight convolutional neural network to reduce improve speed. In this article, depthwise separable residual module is constructed combining convolution non-bottleneck-free module, structure used replace VGG backbone SSD feature extraction parameter quantity At same time, kernels 1 × 3 standard adding 1, respectively, obtain multiple graphs corresponding SSD, established integrating information graphs. self-built dataset complex scenes comparative experiments; experimental results verify effectiveness superiority method. tested video performance model, deployed Android platform scalability model.

Language: Английский

Citations

73
Dynamic Gesture Recognition Using Surface EMG Signals Based on Multi-Stream Residual Network DOI Creative Commons
Zhiwen Yang, Du Jiang, Ying Sun

et al.

Frontiers in Bioengineering and Biotechnology, Journal Year: 2021, Volume and Issue: 9

Published: Oct. 22, 2021

Gesture recognition technology is widely used in the flexible and precise control of manipulators assisted medical field. Our MResLSTM algorithm can effectively perform dynamic gesture recognition. The result surface EMG signal decoding applied to controller, which improve fluency artificial hand control. Much current research using sEMG has focused on static gestures. In addition, accuracy depends extraction selection features. However, Static cannot meet requirements natural human-computer interaction dexterous manipulators. Therefore, a multi-stream residual network (MResLSTM) proposed for movement This study aims stability Simultaneously, it also advance smooth Manipulator. We combine model convolutional short-term memory into unified framework. architecture extracts spatiotemporal features from two aspects: global deep, combines feature fusion retain essential information. strategy pointwise group convolution channel shuffle reduce number calculations. A dataset constructed containing six gestures training. experimental results show that same model, effect acceleration better than only signal. approach obtains competitive performance our with accuracies 93.52%, achieving state-of-the-art 89.65% precision Ninapro DB1 dataset. bionic calculation method realize continuity flexibility manipulator

Language: Английский

Citations

94
Improved single shot multibox detector target detection method based on deep feature fusion DOI
Dongxu Bai, Ying Sun, Bo Tao

et al.

Concurrency and Computation Practice and Experience, Journal Year: 2021, Volume and Issue: 34(4)

Published: Sept. 16, 2021

Summary The feature layers of different in the single shot multibox detector (SSD) are independently used as input classification network, so it is easy to detect same object. This article proposes an improved SSD model based on deep fusion. In algorithm, fusion between target detection layer and its adjacent used, including convolution kernels pooling sizes, down‐sampling low‐level features up‐sampling deconvolution high‐level features. network by combining frame recommendation strategy algorithm regression algorithm. experimental results show that improves accuracy rate target, effect more obvious for relatively small‐scale target.

Language: Английский

Citations

70
Grasping posture of humanoid manipulator based on target shape analysis and force closure DOI Creative Commons
Ying Liu, Du Jiang, Bo Tao

et al.

Alexandria Engineering Journal, Journal Year: 2021, Volume and Issue: 61(5), P. 3959 - 3969

Published: Sept. 21, 2021

With the diversity of manipulator grasping methods and complexity unstructured environment, planning target object is very complicated. However, external factor for primarily shape object. Therefore, it great significance to establish a system based on analysis. This paper proposes method determining posture analysis force closure. The irregular or complex objects are reduced combination some basic shapes. 3D data points split into blocks each part fitted sphere, cylinder rectangle by best-fit algorithm. allows be determined quickly. characteristics analyzed surface described qualitatively means superellipse. simplified according stable condition spiral space defines quality manipulator. best obtained evaluating indicators. eliminates training processes, reduces robotic highly universal.

Language: Английский

Citations

69
Self-adjusting force/bit blending control based on quantitative factor-scale factor fuzzy-PID bit control DOI Creative Commons
Juntong Yun, Ying Sun,

Cuiqiao Li

et al.

Alexandria Engineering Journal, Journal Year: 2021, Volume and Issue: 61(6), P. 4389 - 4397

Published: Oct. 12, 2021

In order to improve the operation performance and environmental adaptability of manipulator, it is essential control position force manipulator simultaneously. This paper proposed a hybrid method auto-dynamic bit based on realize flexible grasping in unknown environment. The constraint conditions types force/bit are analyzed firstly, dynamic model when contacting with environment established. Then, an improved controller conventional fuzzy-PID strategy introduced by combining estimation impedance control. Finally, verified simulation experiments. results show that designed has relatively ideal precision higher trajectory tracking ability, better compliance.

Language: Английский

Citations

63