检索结果-南通市图书馆

Moisture influence in emerging neuromorphic device

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Frontiers of physics 2023年第5期18卷 133-142页

作者： Wenhua Wang Guangdong Zhou College of Artificial Intelligence Southwest UniversityChongqing 400715China

Conduction filament formation,redox reaction,and mobile ion migration in solid electrolytes underpin the memristive devices,all of which are partially influenced or fully dominated by the moisture.The moisture-based physical-chemistry mechanism provides an electric tunable method to create enough dissociate conductance states for neuromorphic computing,but overconcentration moisture will corrode electrode and then causes device invalidation.This perspective goal is that surveys the moisture-dependency of dynamic at interfaces or/and switching function layer,clarifies the bottlenecks that the memristive device facing in terms of water molecule-related reaction,and gives the possible solutions.

关键词： memristor moisture redox reaction oxide interface engineering

HAPNet: a head-aware pedestrian detection network associated with the affinity field

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Science China(Information Sciences) 2022年第6期65卷 17-32页

作者： Jiali DING Tie LIU Yun ZHAO Zejian YUAN Yuanyuan SHANG college of Information Engineering Capital Normal University College of Artificial Intelligence Xi'an Jiaotong University

Pedestrian detection has made great progress with the rapid development of deep learning, but pedestrian detection under occlusion remains a challenge. To solve the occlusion problem, some research endeavors have been carried out based on visible parts, such as the body, but there is still substantial room for improvement due to the introduction of additional calculation. Aiming to improve the detection performance under occlusion, this paper proposes a head-aware pedestrian detection network(HAPNet) by using the inherent structural relationship between the human body and head. The postprocessing stage is redesigned to include a scoring module and an augmented non-maximum suppression(NMS) algorithm.Specifically, HAPNet detects head and body simultaneously through different layers of a feature map. We then propose a head-side affinity model, which can represent the association between the head and body sides. Detection and affinity prediction tasks are implemented through different branches of HAPNet. In the scoring module, the head and body detection scores are fused to match the head-body pairs to improve the detection performance. On this basis, an enhanced NMS algorithm is proposed, which achieves a good balance between reducing false positives and missing detection. The experimental results verify the effectiveness of this method in pedestrian detection under occlusion.

关键词： pedestrian detection head detection head-aware pedestrian network affinity module occlusion

Optimization-inspired manual architecture design and neural architecture search

同方期刊数据库评论

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Science China(Information Sciences) 2023年第11期66卷 100-112页

作者： Yibo YANG Zhengyang SHEN Huan LI Zhouchen LIN JD Explore Academy School of Mathematical Sciences Peking University Institute of Robotics and Automatic Information Systems College of Artificial Intelligence Nankai University Key Lab of General artificial intelligence School of Intelligence Science and Technology Peking University Institute for artificial intelligence Peking University Pazhou Lab

Neural architecture has been a research focus in recent years due to its importance in deciding the performance of deep networks. Representative ones include a residual network(Res Net) with skip connections and a dense network(Dense Net) with dense connections. However, a theoretical guidance for manual architecture design and neural architecture search(NAS) is still lacking. In this paper, we propose a manual architecture design framework, which is inspired by optimization algorithms. It is based on the conjecture that an optimization algorithm with a good convergence rate may imply a neural architecture with good performance. Concretely, we prove under certain conditions that forward propagation in a deep neural network is equivalent to the iterative optimization procedure of the gradient descent algorithm minimizing a cost function. Inspired by this correspondence, we derive neural architectures from fast optimization algorithms,including the heavy ball algorithm and Nesterov’s accelerated gradient descent algorithm. Surprisingly, we find that we can deem the Res Net and Dense Net as special cases of the optimization-inspired architectures.These architectures offer not only theoretical guidance, but also good performances in image recognition on multiple datasets, including CIFAR-10, CIFAR-100, and Image Net. Moreover, we show that our method is also useful for NAS by offering a good initial search point or guiding the search space.

关键词： deep neural network manual architecture design neural architecture search image recognition optimization algorithms learning-based optimization

Intelligent sorting method for assembly line based on visual positioning and robotic arm model predictive control

同方期刊数据库评论

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

International Journal of Agricultural and Biological Engineering 2023年第4期16卷 206-214页

作者： Ruining Zhang Wei Lu Xingliang Jian Hui Luo College of Artificial Intelligence Nanjing Agricultural UniversityNanjing 210095China

The existing steering device in the fruit and vegetable packaging assembly line cannot adjust the attitude of lettuce to a unified attitude,affecting the input and packaging process of the packaging machine.This study proposes an intelligent assembly line sorting method based on the visual positioning and model predictive control of a robotic arm.First,lightweight improvement based on the YOLOv5 is realized,the lettuce stalk in the background of the conveyor belt is promptly identified,the image of the lettuce stalk in the anchor box area is processed,and the edge contour point set is determined to extract the pixel coordinates of the optimal grasp point and mirror inclination angle of the lettuce.For the intelligent assembly line system,a robot arm kinematics model is constructed and the robot kinematics inverse solutions are calculated.Additionally,the lettuce movement speeds are dynamically measured by the vision system.A combination of the model prediction control,dynamic tracking,and rapid sorting of the lettuce by the robot claw is realized.The results show that the average detection time of a single frame image in the visual positioning part is 0.014 s,which is reduced by 50%;the accuracy and recall are 98%and 95%,respectively.The detection time is significantly reduced by ensuring accuracy.Within the current speed range of the packaging assembly line conveyor belt,the manipulator can grasp lettuce at different speeds stably and fast;the average axial error,average radial error,and adjusted average inclination angle error are 0.71 cm,1.02 cm,and 3.79°,respectively,verifying the high efficiency and stability of the model.The proposed method of this study enables application in the intelligent sorting operation of industrial assembly lines.

关键词： YOLOv5 deep learning image recognition model predictive control intelligent assembly line

Fine-grained detection of caged-hen head states using adaptive Brightness Adjustment in combination with Convolutional Neural Networks

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

International Journal of Agricultural and Biological Engineering 2023年第3期16卷 208-216页

作者： Jia Chen Qi’an Ding Wen Yao Mingxia Shen Longshen Liu College of Artificial Intelligence Nanjing Agricultural UniversityNanjing 210095China

Timely identification and tracking of abnormal hens in stacked cages are of great significance for precision treatment and the elimination of sick individuals.The head features of the caged-hens are used to overcome observation difficulties caused by the cage and feathers blocking,but it is still hard to identify similar head states.To solve this problem,the fine-grained detection of caged-hens head states was developed using adaptive Brightness Adjustment in combination with Convolutional Neural Networks(FBA-CNN).Grid Region-based CNN(R-CNN),a convolution neural network(CNN),was optimized with the Squeeze-and-Excitation(SE)and Depthwise Over-parameterized Convolutional(DO-Conv)to detect layer heads from cages and to accurately cut them as single-head images.The brightness of each single-head image was adjusted adaptively and classified through the deep convolution neural network based on SE-Resnet50.Finally,we returned to the original image to realize multi-target detection with coordinate mapping.The results showed that the AP@0.5 of layer head detection using the optimized Grid R-CNN was 0.947,the accuracy of classification with SE-Resnet50 was 0.749,the F1 score was 0.637,and the mAP@0.5 of FBA-CNN was 0.846.In summary,this automated method can accurately identify different layer head states in layer cages to provide a basis for follow-up studies of abnormal behavior including dyspnea and cachexia.

关键词： Grid R-CNN squeeze-and-excitation Depthwise Over-parameterized Convolutional adaptive brightness adjustment fine-grained detection

Surface defect detection competition with a bio-inspired vision sensor

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

National Science Review 2023年第6期10卷 31-34页

作者： Anjing Guo Jupo Ma Renwei Dian Fuyan Ma Jinjian Wu Shutao Li School of Robotics Hunan University School of artificial intelligence Xidian University Pazhou Laboratory (Huangpu) college of Electrical and Information Engineering Hunan University Key Laboratory of Visual Perception and artificial intelligence of Hunan Province

INTRODUCTION Surface defects are inevitable during the product manufacturing process,such as impurities,scratches and dirt.D efects not only reduce the aesthetics and comfort of products,but may also degrade the functional performance.Manufacturers are very concerned about inadequate defect detection because of potential economic and reputation losses.Thus,it is significant to study automatic defect detection methods.Existing machine vision-based defect detection mostly adopts a traditional CCD or CMOS vision sensor (VS) for imaging.However,in real-world production lines,due to dynamic environments,traditional VS encounters difficulties in meeting the demands of defect detection,especially in modern high-precision product manufacturing.

关键词：

Microfluidic bioscaffolds for regenerative engineering

同方期刊数据库评论

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Engineered Regeneration

Engineered Regeneration 2022年第1期3卷 110-120页

作者： Wantao Zhuge Han Liu Wuyuan Wang Jie Wang College of Artificial Intelligence Nanjing Agricultural UniversityNanjing 210031China

The construction of biological scaffolds in vitro plays an important role in regenerative engineering,which has always been the focus of research.The application of scaffold materials combined with cells/bioactive factors is considered to have great potential for tissue regeneration.Various strategies have been used for constructing biological scaffolds,including electrospinning/spraying,direct laser writing,solvent casting,and microfluidic technology.Among them,considering the advantages of the safety,low cost,and highly controllable process-ing characteristics,microfluidic technology has become a promising method for the construction of biological scaffolds.This review overviews the recent research progress of microfluidic bioscaffolds for regenerative engi-neering.Firstly,we introduce some typical natural polymers which are widely used to construct bioscaffolds.Then,the bioscaffolds with different structures templated by microfluidic droplets and fibers are described.Fur-therly,we talk about the different applications of microfluidic bioscaffolds in wound healing,drug delivery,bone regeneration,and nerve regeneration.Finally,the challenges and future prospects of the development of microfluidic bioscaffolds for regenerative engineering are discussed.

关键词： Regenerative engineering Microfluidics Bioscaffold Polymer

Analysis of CLARANS Algorithm for Weather Data Based on Spark

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Computers, Materials & Continua 2023年第8期76卷 2427-2441页

作者： Jiahao Zhang Honglin Wang College of Artificial Intelligence Nanjing University of Information Science and TechnologyNanjing210044China

With the rapid development of technology,processing the explosive growth of meteorological data on traditional standalone computing has become increasingly time-consuming,which cannot meet the demands of scientific research and business.Therefore,this paper proposes the implementation of the parallel Clustering Large Application based upon RANdomized Search(CLARANS)clustering algorithm on the Spark cloud computing platformto cluster China’s climate regions usingmeteorological data from1988 to 2018.The aim is to address the challenge of applying clustering algorithms to large datasets.In this paper,the morphological similarity distance is adopted as the similarity measurement standard instead of Euclidean distance,which improves clustering accuracy.Furthermore,the issue of local optima caused by an improper selection of initial clustering centers is addressed by utilizing the max-distance criterion.Compared to the k-means clustering algorithm already implemented in the Spark platform,the proposed algorithm has strong robustness,can reduce the interference of outliers in the dataset on clustering results,and has higher parallel performance than the frequently used serial algorithms,thus improving the efficiency of big data analysis.This experiment compares the clustered centroid data with the annual average meteorological data of representative cities in the five typical meteorological regions that exist in China,and the results show that the clustering results are in good agreement with the meteorological data obtained from the National Meteorological Science Data Center.This algorithm has a positive effect on the clustering analysis of massive meteorological data and deserves attention in scientific research activities.

关键词： Clustering analysis cloud computing platform parallel algorithm

SC-Net:A New U-Net Network for Hippocampus Segmentation

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Intelligent Automation & Soft Computing 2023年第9期37卷 3179-3191页

作者： Xinyi Xiao Dongbo Pan Jianjun Yuan College of Artificial Intelligence Southwest UniversityChongqing400715China

Neurological disorders like Alzheimer’s disease have a significant impact on the lives and health of the elderly as the aging population con-tinues to grow.Doctors can achieve effective prevention and treatment of Alzheimer’s disease according to the morphological volume of hippocam-pus.General segmentation techniques frequently fail to produce satisfactory results due to hippocampus’s small size,complex structure,and fuzzy edges.We develop a new SC-Net model using complete brain MRI images to achieve high-precision segmentation of hippocampal structures.The proposed network improves the accuracy of hippocampal structural segmentation by retaining the original location information of the hippocampus.Extensive experimental results demonstrate that the proposed SC-Net model is signif-icantly better than other models,and reaches a Dice similarity coefficient of 0.885 on Alzheimer’s Disease Neuroimaging Initiative(ADNI)dataset.

关键词： SC-Net hippocampus brain MRI images image segmentation