Multi-Scale Feature Fusion Convolutional Neural Network for Indoor Small Target Detection

Frontiers in Neurorobotics, Год журнала: 2022, Номер 16

Опубликована: Май 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.

Язык: Английский

---

Low-Illumination Image Enhancement Algorithm Based on Improved Multi-Scale Retinex and ABC Algorithm Optimization

Frontiers in Bioengineering and Biotechnology, Год журнала: 2022, Номер 10

Опубликована: Апрель 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.

Язык: Английский

---

Real-Time Target Detection Method Based on Lightweight Convolutional Neural Network

Frontiers in Bioengineering and Biotechnology, Год журнала: 2022, Номер 10

Опубликована: Авг. 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.

Язык: Английский

---

Lightweight Network for Corn Leaf Disease Identification Based on Improved YOLO v8s

Agriculture, Год журнала: 2024, Номер 14(2), С. 220 - 220

Опубликована: Янв. 29, 2024

This research tackles the intricate challenges of detecting densely distributed maize leaf diseases and constraints inherent in YOLO-based detection algorithms. It introduces GhostNet_Triplet_YOLOv8s algorithm, enhancing YOLO v8s by integrating lightweight GhostNet (Ghost Convolutional Neural Network) structure, which replaces backbone. adaptation involves swapping head’s C2f (Coarse-to-Fine) Conv (Convolutional) modules with C3 Ghost GhostNet, simplifying model architecture while significantly amplifying speed. Additionally, a attention mechanism, Triplet Attention, is incorporated to refine accuracy identifying post-neck layer output precisely define features within disease-affected areas. By introducing ECIoU_Loss (EfficiCLoss Loss) function, replacing original CIoU_Loss, algorithm effectively mitigates issues associated aspect ratio penalties, resulting marked improvements recognition convergence rates. The experimental outcomes display promising metrics precision rate 87.50%, recall 87.70%, an [email protected] 91.40% all compact size 11.20 MB. In comparison v8s, this approach achieves 0.3% increase mean average (mAP), reduces 50.2%, decreases FLOPs 43.1%, ensuring swift accurate disease identification optimizing memory usage. Furthermore, practical deployment trained on WeChat developer mini-program underscores its utility, enabling real-time fields aid timely agricultural decision-making prevention strategies.

Язык: Английский

---

A Tandem Robotic Arm Inverse Kinematic Solution Based on an Improved Particle Swarm Algorithm

Frontiers in Bioengineering and Biotechnology, Год журнала: 2022, Номер 10

Опубликована: Май 19, 2022

The analysis of robot inverse kinematic solutions is the basis control and path planning, great importance for research. Due to limitations analytical geometric methods, intelligent algorithms are more advantageous because they can obtain approximate directly from robot's positive equations, saving a large number computational steps. Particle Swarm Algorithm (PSO), as one algorithms, widely used due its simple principle excellent performance. In this paper, we propose an improved particle swarm algorithm kinematics solving. Since setting weights affects global local search ability algorithm, paper proposes adaptive weight adjustment strategy improving ability. Considering running time condition based on limit joints, introduces position coefficient k in velocity factor. Meanwhile, exponential product form modeling method (POE) spinor theory chosen. Compared with traditional DH method, approach describes motion rigid body whole avoids singularities that arise when described by coordinate system. order illustrate advantages terms accuracy, time, convergence adaptability, three experiments were conducted general six-degree-of-freedom industrial robotic arm, PUMA560 arm seven-degree-of-freedom research objects. all experiments, parameters range joint angles, initial attitude end-effector given, impact point set verify whether angles found reach specified positions. Experiments 2 3, proposed compared (PSO) quantum (QPSO) direction solving operation convergence. results show other two ensure higher accuracy orientation end-effector. error 0 maximum 1.29 × 10-8. while minimum -1.64 10-5 -4.03 10-6. has shorter algorithms. last computing 0.31851 0.30004s respectively, shortest 0.33359 0.30521s respectively. convergence, achieve faster stable than After changing experimental subjects, still maintains which indicates applicable certain potential multi-arm solution. This provides new way thinking solution problem.

Язык: Английский

---

Time Optimal Trajectory Planing Based on Improved Sparrow Search Algorithm

Frontiers in Bioengineering and Biotechnology, Год журнала: 2022, Номер 10

Опубликована: Март 22, 2022

Complete trajectory planning includes path planning, inverse solution solving and optimization. In this paper, a highly smooth time-saving approach to is obtained by improving the kinematic optimization algorithms for time-optimal problem. By partitioning joint space, paper obtains an calculation based on of saving 40% kinematics time. This means that large number computational resources can be saved in planning. addition, improved sparrow search algorithm (SSA) proposed complete trajectory. A Tent chaotic mapping was used optimize way generating initial populations. The further combining it with adaptive step factor. experiments demonstrated performance SSA. robot’s optimized time algorithm. Experimental results show method improve convergence speed global capability ensure trajectories.

Язык: Английский

---

Path Planning Optimization of Intelligent Vehicle Based on Improved Genetic and Ant Colony Hybrid Algorithm

Frontiers in Bioengineering and Biotechnology, Год журнала: 2022, Номер 10

Опубликована: Июль 1, 2022

Intelligent vehicles were widely used in logistics handling, agriculture, medical service, industrial production, and other industries, but they often not smooth enough planning the path, number of turns was large, resulting high energy consumption. Aiming at unsmooth path problem four-wheel intelligent vehicle algorithm, this article proposed an improved genetic ant colony hybrid physical model established. This first optimization algorithm about heuristic function with adaptive change evaporation factor. Then, it on fitness function, adjustment crossover factor, mutation Last, addition a deletion operator, adoption elite retention strategy, suboptimal solutions obtained from to obtain optimized new populations. The simulation environment for is windows 10, processor Intel Core i5-5257U, running memory 4GB, compilation MATLAB2018b, samples 50, maximum iterations 100, initial population size 200, 50. Simulation experiments show that effective. Compared traditional reduced by 46% average 75% simple grid. 47% 21% complex works better reduce maps.

Язык: Английский

---

Improved Multi-Stream Convolutional Block Attention Module for sEMG-Based Gesture Recognition

Frontiers in Bioengineering and Biotechnology, Год журнала: 2022, Номер 10

Опубликована: Июнь 7, 2022

As a key technology for the non-invasive human-machine interface that has received much attention in industry and academia, surface EMG (sEMG) signals display great potential advantages field of collaboration. Currently, gesture recognition based on sEMG suffers from inadequate feature extraction, difficulty distinguishing similar gestures, low accuracy multi-gesture recognition. To solve these problems new network called Multi-stream Convolutional Block Attention Module-Gate Recurrent Unit (MCBAM-GRU) is proposed, which signals. The multi-stream formed by embedding GRU module CBAM. Fusing ACC further improves action experimental results show proposed method obtains excellent performance dataset collected this paper with accuracies 94.1%, achieving advanced 89.7% Ninapro DB1 dataset. system high classifying 52 kinds different delay less than 300 ms, showing terms real-time human-computer interaction flexibility manipulator control.

Язык: Английский

---

Attitude Stabilization Control of Autonomous Underwater Vehicle Based on Decoupling Algorithm and PSO-ADRC

Frontiers in Bioengineering and Biotechnology, Год журнала: 2022, Номер 10

Опубликована: Фев. 28, 2022

Autonomous Underwater Vehicle are widely used in industries, such as marine resource exploitation and fish farming, but they often subject to a large amount of interference which cause poor control stability, while performing their tasks. A decoupling algorithm is proposed single volume-single attitude angle model constructed for the problem severe coupling system six degrees freedom Vehicle. Aiming at complex Active Disturbance Rejection Control (ADRC) adjustment relying on manual experience, PSO-ADRC realize automatic its parameters, improves anti-interference ability accuracy dynamic environment. The method were verified through experiments.

Язык: Английский

---

Grasping Pose Detection for Loose Stacked Object Based on Convolutional Neural Network With Multiple Self-Powered Sensors Information

IEEE Sensors Journal, Год журнала: 2022, Номер 23(18), С. 20619 - 20632

Опубликована: Авг. 5, 2022

There are a variety of objects, random postures and multiple objects stacked in disorganized manner unstructured home applications, which leads to the object grasping posture estimation planning based on machine vision become very complicated. This paper proposes method cluttering pose detection convolutional neural network with self-powered sensors information. Firstly, search strategy for candidate poses 3D point cloud is proposed, single-channel image dataset representing this established by using Bigbird dataset. Secondly, ResNet constructed rank filter single channel captured images bit pose. It also compared three mainstream classification networks, Inception V2, VGG-A LetNet, perception analysis function execution developed under ROS. The effective manipulator scene scattered piles realized results position combined information sensors, other networks. In environment experiment show that superior average success rate ResNet, InceptionV2, VGGA LetNet networks 90.67%, 82.67%, 86.67% 87.33% respectively, verifies effectiveness superiority deep learn-based model proposed paper.

Язык: Английский

---