Article
Environmental Sciences
Guangzhe Si, Ying Xiao, Bin Wei, Leon Bevan Bullock, Yueyue Wang, Xiaodong Wang
Summary: This paper introduces a novel Transformer-based framework, Token-Selective Vision Transformer (TSVT), for fine-grained image classification of marine organisms. The framework utilizes Token-Selective Self-Attention (TSSA) to select important tokens and gradually narrow down to more precise local regions, achieving state-of-the-art performance.
FRONTIERS IN MARINE SCIENCE
(2023)
Article
Physics, Multidisciplinary
Ching-Hsun Tseng, Shin-Jye Lee, Jianan Feng, Shengzhong Mao, Yu-Ping Wu, Jia-Yu Shang, Xiao-Jun Zeng
Summary: This work proposes an efficient and robust backbone, UPANets, which utilizes channel and spatial direction attentions to expand the receptive fields in shallow convolutional layers. Experimental results show that UPANets achieve better performance with fewer resources on CIFAR-{10, 100} than existing state-of-the-art methods.
Article
Computer Science, Artificial Intelligence
Jingkai Zhou, Pichao Wang, Jiasheng Tang, Fan Wang, Qiong Liu, Hao Li, Rong Jin
Summary: Although self-attention is powerful in modeling long-range dependencies, the performance of local self-attention (LSA) is similar to depth-wise convolution. To clarify the differences and limitations, a comprehensive investigation on LSA and its counterparts was conducted. The study finds that attention generation and application, including relative position embedding and neighboring filter application, are key factors. Based on these findings, an enhanced version called ELSA is proposed, which introduces Hadamard attention and the ghost head. Experiments demonstrate the effectiveness of ELSA in various tasks.
INTERNATIONAL JOURNAL OF COMPUTER VISION
(2023)
Article
Computer Science, Artificial Intelligence
Xujia Hou, Feihu Zhang, Dhiraj Gulati, Tingfeng Tan, Wei Zhang
Summary: Event cameras have the advantages of low delay, high dynamic range, and no motion blur, but encounter obstacles due to their unique data representation. This study proposes a network called E2VIDX based on the image reconstruction algorithm for event cameras, which achieves better feature fusion and reduces the network model size. A new loss function is also introduced. Experimental results show significant improvements compared to existing methods, with increased Structural Similarity (SSIM) and decreased Learned Perceptual Image Patch Similarity (LPIPS) and Mean Squared Error (MSE). The proposed method is also evaluated in image classification, object detection, and instance segmentation, demonstrating its effectiveness in applying existing vision algorithms.
FRONTIERS IN NEUROROBOTICS
(2023)
Article
Computer Science, Artificial Intelligence
Kaibo Duan, Shi Bao, Zhiqiang Liu, Shaodong Cui
Summary: Pollen identification has broad applications in various fields, and pollen allergy is a common and frequent disease. Accurate and rapid identification of pollen species under the electron microscope can help with pollen forecast and treatment. In this study, a new Vision Transformer pipeline for image classification is proposed, which achieves CNN-equivalent performance on the pollen dataset with fewer model parameters and training time.
NEURAL COMPUTING & APPLICATIONS
(2023)
Article
Computer Science, Information Systems
Yupeng Song, Fazhi He, Yanan Liu
Summary: In this paper, a non-invasive attention learning branch structure is proposed to enhance the discrimination capacity of convolutional neural networks (CNNs) for complex tree species classification. By adding an attention learning branch to the typical convolutional blocks, key features of leaf images can be extracted and integrated into the classification process, resulting in remarkable improvement in performance.
MULTIMEDIA TOOLS AND APPLICATIONS
(2022)
Article
Environmental Sciences
Reenul Reedha, Eric Dericquebourg, Raphael Canals, Adel Hafiane
Summary: This paper explores the potential of attention-based deep networks (ViT) in weed and crop recognition using drone systems. It demonstrates that ViT models outperform state-of-the-art models with small labeled training datasets, showing promise in a wide range of remote sensing image analysis tasks.
Article
Environmental Sciences
Yakoub Bazi, Laila Bashmal, Mohamad M. Al Rahhal, Reham Al Dayil, Naif Al Ajlan
Summary: This paper proposes a remote-sensing scene-classification method based on vision transformers, which utilize multihead attention mechanisms to establish long-range contextual relationships between pixels in images. The approach involves dividing images into patches, converting them into sequences, and applying data augmentation techniques for improved classification performance. The study also demonstrates the efficacy of compressing the network by pruning half of the layers while maintaining competitive classification accuracies.
Article
Computer Science, Artificial Intelligence
Qi Wang, JianJun Wang, Hongyu Deng, Xue Wu, Yazhou Wang, Gefei Hao
Summary: Fine-grained visual classification is a difficult task due to the large inter-class variances and small intra-class variances. Existing approaches using CNN-based networks as feature extractors fail to locate the important parts. In this paper, we propose an attention aggregating transformer (AATrans) based on the ViT model to better capture minor differences among images and improve performance. We introduce the core attention aggregator (CAA) and the information entropy selector (IES) to enhance information sharing and select discriminative parts. Extensive experiments demonstrate the state-of-the-art performance of our proposed model on mainstream datasets.
PATTERN RECOGNITION
(2023)
Article
Computer Science, Information Systems
Jing Wang, Yehao Li, Yingwei Pan, Ting Yao, Jinhui Tang, Tao Mei
Summary: This paper introduces a new design to explore the interdependencies between attention histories and emphasize the focus of each attention in image captioning. By memorizing contextual attention and extracting principal components from each attention, the proposed CoSA-Net achieves superior performance improvement.
SCIENCE CHINA-INFORMATION SCIENCES
(2022)
Article
Environmental Sciences
Jiping Zhai, Lu Han, Ying Xiao, Mai Yan, Yueyue Wang, Xiaodong Wang
Summary: Accurately classifying marine fish species is important for marine ecosystem investigations. Previously used methods were labor-intensive, but computer vision approaches offer advantages such as long-term, non-destructive, and low-cost classification. This paper proposes a novel attention network model that enhances feature extraction from few-shot fine-grained fish images, improving classification accuracy.
FRONTIERS IN MARINE SCIENCE
(2023)
Article
Computer Science, Information Systems
Jing Li, Xueping Luo
Summary: A weather-image-classification model is proposed that combines a VIT and dual augmented attention module to overcome the limitations of traditional deeplearning methods in feature extraction, recognition accuracy, and the limited types of weather phenomena in datasets. The model utilizes a pre-trained VIT to acquire basic semantic features and incorporates dual augmented attention with convolutional self-attention and Atrous self-attention modules to capture low-level and high-level deep-image semantics. Experimental validation on standard weather-image datasets demonstrates the model's effectiveness, achieving higher F1 scores than recent deep-learning models in the comparison.
Article
Computer Science, Artificial Intelligence
Peisong Wang, Fanrong Li, Gang Li, Jian Cheng
Summary: In this article, the advantages of extremely sparse networks with binary connections for image classification through software-hardware codesign are investigated. A binary augmented extremely pruning method is proposed to achieve high sparsity with minimal accuracy degradation, and a hardware architecture based on the resulting sparse and binary networks is designed to explore the benefits of extreme sparsity with negligible consumption. Experiments on ImageNet classification and FPGA demonstrate a significant tradeoff between accuracy and efficiency in the proposed software-hardware architecture.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Chemistry, Multidisciplinary
Chang Chen, Huaixiang Zhang
Summary: Image classification is crucial in computer vision with a wide range of applications. This study proposes a novel pooling operation, named Binary Pooling, which combines Global Average Pooling (GAP) and Global Max Pooling (GMP) to extract more comprehensive image features. By applying dilation operations and pointwise convolutions, the extraction of image features is further enhanced. Experimental results show that integrating this attention block into ResNet18/50 models improves accuracy on ImageNet.
APPLIED SCIENCES-BASEL
(2023)
Article
Computer Science, Artificial Intelligence
Siqi Liu, Jiangshu Wei, Gang Liu, Bei Zhou
Summary: This paper proposes a hybrid model combining CNN and Transformer to optimize their shortcomings in capturing global feature representation and deteriorating local feature details. The model achieves accurate localization of features and efficient capture of long-range relationships on a large receptive field. Experimental results demonstrate the excellent performance of the proposed model on cifar10, cifar100, and birds400 datasets with fewer model parameters.
PEERJ COMPUTER SCIENCE
(2023)
Article
Engineering, Electrical & Electronic
Bangyan Zeng, Xianghong Zhang, Changsong Gao, Yi Zou, Xipeng Yu, Qian Yang, Tailiang Guo, Huipeng Chen
Summary: With high efficiency and low energy consumption, bio-inspired artificial neuromorphic systems have attracted tremendous attention. In this work, an artificial optoelectronic neural device based on an Ag/MXene/SiO2/Si structure is demonstrated. By introducing an oxidized MXene (O-MXene) layer, photoelectric integration can be realized on a single device, which efficiently accelerates the firing behavior of neurons. Additionally, a 64 x 64 sensing array based on optoelectronic neurons is successfully designed and demonstrated for signal recognition and sharpening.
IEEE TRANSACTIONS ON ELECTRON DEVICES
(2023)
Article
Materials Science, Multidisciplinary
Chaomin Mao, Songman Ju, Jinping Zheng, Yueting Zheng, Zhongwei Xu, Lihua Lin, Hailong Hu, Kaiyu Yang, Tailiang Guo, Fushan Li
Summary: This study fabricates high-resolution perovskite quantum dot LEDs (PQLEDs) using nanoimprint technique, which can achieve massive pixel density displays with high color purity, efficiency, and wide color gamut.
ADVANCED OPTICAL MATERIALS
(2023)
Article
Engineering, Electrical & Electronic
Jianpu Lin, Hongxing Xie, Yun Ye, Tailiang Guo
Summary: This paper presents a photoluminescent quantum dot (QD) printing paste for light guide dots array used in the backlight of liquid crystal display (LCD). Different solvents were used to prepare and characterize the QD printing paste, which was then enhanced with an oxygen inhibitor called triphenyl phosphite (TPP) to improve its stability. The QD printing paste was used in screen printing to fabricate a QD light guide plate (LGP) with silicon dioxide (SiO2) light scattering particles. The QDs LGP achieved a color gamut of 131.6%NTSC due to light scattering interaction among blue LED, red QDs, green QDs, and SiO2 particles. The study shows that the QDs-chloroform printing paste has uniform light output and the highest photoluminescent peak intensity, making it a promising option for the LCD backlight with its simple process, high production efficiency, and low cost.
JOURNAL OF THE SOCIETY FOR INFORMATION DISPLAY
(2023)
Article
Engineering, Electrical & Electronic
Wenwen Wang, Yongai Zhang, Chaoxing Wu, Qun Yan, Tailiang Guo, Xiongtu Zhou
Summary: A microfabrication method based on bifocal microlens arrays (MLAs) is proposed to improve the depth of field (DOF) in integral imaging display. The bifocal MLAs, fabricated using two-step photolithography and thermal reflow, allow high spatial resolution and accurate depth estimation by using microlenses of different focal lengths. Hexagonally packaged bifocal MLAs show extended DOF from 0.004 to 4.908 mm for 57.6 μm lens diameter, with corresponding object distance ranges from 0.125 to 0.165 mm.
JOURNAL OF THE SOCIETY FOR INFORMATION DISPLAY
(2023)
Article
Engineering, Electrical & Electronic
Yao Li, Haonan Jiang, Yinguo Yan, Yongzhen Liu, Ziping Zhou, Enguo Chen, Yun Ye, Sheng Xu, Qun Yan, Tailiang Guo
Summary: This paper presents a pico-projector design based on micro-LED (mu LED) light sources, which includes a highly integrated microlens array to improve system efficiency and a set of four-piece spherical lens group to achieve compact size. The simulation results show that the mu LED equipped with the microlens array can increase light energy utilization by about 3.5 times and reduce the size of the pico-projector to 30.18 mm(3). The designed pico-projector has high performance and compact size with great potential for future applications.
JOURNAL OF THE SOCIETY FOR INFORMATION DISPLAY
(2023)
Article
Engineering, Electrical & Electronic
Aochen Du, Wenxiao Zhao, Yun Ye, Enguo Chen, Sheng Xu, Tailiang Guo
Summary: In this paper, Cs(Pb,Sb)Br-3 PQDs@glasses were successfully prepared by traditional methods. The optical characterization showed that Cs(Pb0.7Sb0.3)Br-3 PQDs@glasses have a emission peak at 518 nm and a full width at half maximum of 20 nm, with a photoluminescence quantum yield (PLQY) of 58%. The stability and potential applications of the PQDs@glass were further studied through thermal analysis and combining with ceramic phosphors.
JOURNAL OF THE SOCIETY FOR INFORMATION DISPLAY
(2023)
Review
Engineering, Aerospace
Min Qian, Yi Zhang, Xiaojun Mao, Yang Gao, Xiaoyang Xuan, Min Wu, Yueping Niu, Shangqing Gong
Summary: Flexibility and lightweight are important research topics in space science and technology. However, most photoelectronic devices on spacecraft are currently rigid, which is not suitable for the space environment with irradiations and thermal cycling. This review presents the synthesis of space-durable photoelectronic materials, the fabrication of flexible devices, and the investigation of irradiation mechanisms. These advancements lead the development of flexible and lightweight space science and technology.
PROGRESS IN AEROSPACE SCIENCES
(2023)
Correction
Engineering, Electrical & Electronic
Xiang Zhang, Anlan Chen, Tao Yang, Junhu Cai, Yuanyuan Ye, Enguo Chen, Sheng Xu, Yun Ye, Jie Sun, Qun Yan, Tailiang Guo
IEEE PHOTONICS JOURNAL
(2023)
Article
Materials Science, Multidisciplinary
Lujian Liu, Qizhen Chen, Huaan Zeng, Liuting Shan, Chuanbin An, Bingyong Zhuang, Huipeng Chen, Tailiang Guo, Wenping Hu
Summary: Artificial synaptic devices are important for artificial neural networks. In this study, a quantum-dot light-emitting synaptic transistor capable of dual output of optoelectronic signals is demonstrated. The device can exhibit dual responses of electrical and optical signals and successfully simulate key synaptic functions, showing potential for the development of neuromorphic computing.
ADVANCED MATERIALS TECHNOLOGIES
(2023)
Article
Chemistry, Multidisciplinary
Kuan Ju, Yue Miao, Qi Li, Yabin Yan, Yang Gao
Summary: In this study, the laser direct writing method was used to successfully pyrolyze MnCO3/carboxymethylcellulose (CMC) precursors into MnO2/carbonized CMC (LP-MnO2/CCMC) in a one-step and mask-free manner to meet the requirements of environmentally friendly, simple, and effective material synthesis. CMC was utilized as a combustion-supporting agent to promote the conversion of MnCO3 into MnO2. The selected materials have advantages such as solubility of MnCO3, eco-friendly and soluble carbonaceous nature of CMC, and easy removal of redundant precursor using deionized water. The electrochemical performance of LP-MnO2/CCMC(R1) and LP-MnO2/CCMC(R1/5) composites with different mass ratios of MnCO3 and CMC was investigated. LP-MnO2/CCMC(R1/5)-based electrode exhibited high specific capacitance (74.2 F/g at 0.1 A/g) and good electrical durability for 1000 charging-discharging cycles. The sandwich-like supercapacitor assembled with LP-MnO2/CCMC(R1/5) electrodes showed a maximum specific capacitance of 49.7 F/g at 0.1 A/g. Furthermore, the LP-MnO2/CCMC(R1/5)-based energy supply system successfully powered a light-emitting diode, demonstrating the great potential of LP-MnO2/CCMC(R1/5)-based supercapacitors for power devices.
Article
Instruments & Instrumentation
Zhenglin Li, Taotao Ding, Biao Xiao, Yang Gao, Yanxun Xiang, Fuzhen Xuan
Summary: In this study, a multifunctional sensing layer (MSL) integrated with flexible multi-functional sensing units and stretchable interconnectors was developed using a laser-microfabrication (LMF) method. The MSL showed high sensitivity, durability, and temperature coefficient, making it suitable for various applications in structural health monitoring (SHM).
SMART MATERIALS AND STRUCTURES
(2023)
Article
Chemistry, Multidisciplinary
Aochen Du, Debing Shen, Wenxiao Zhao, Yongzhen Liu, Xinzhi Qin, Zexi Lin, Yun Ye, Enguo Chen, Sheng Xu, Tailiang Guo
Summary: In this study, the electronic structure, phonon dispersion, and vibrational properties of BA(2)PbI(4) crystals were revealed using density functional theory. The stability diagram of formation enthalpy was calculated, and the crystal structure was characterized through Rietveld refinement. The excellent photoelectric properties of BA(2)PbI(4) crystals were confirmed through the design of a contactless fixed-point lighting device and the study of photoelectrochemical properties.
Article
Chemistry, Multidisciplinary
Lei Wang, Xinqi Yao, Shuaishuai Yuan, Yang Gao, Ruhang Zhang, Xinhai Yu, Shan-Tung Tu, Shijian Chen
Summary: An ultra-high performance humidity sensor based on CuO/Ti3C2TX MXene was investigated in this study. The moisture-sensitive material was fabricated using a self-assembly method and characterized using various techniques. The results showed that the humidity sensor exhibited high sensitivity, short response and recovery time, low hysteresis, and good repeatability. The CuO/Ti3C2TX sensor demonstrated great potential in various applications such as respiration rate monitoring and environmental detection.
Proceedings Paper
Optics
Zhenyu Zeng, Hongxing Xie, Yaqian Zheng, Sijie Li, Yun Ye, TaiLiang Guo
Summary: In this study, quantum dots were added into masterbatches to prepare composite components with diffusion and color conversion functions. The resulting QD color masterbatch exhibited uniform dispersion, controllable concentration, and good luminescence performance. This study demonstrates the potential application of QD masterbatches in high-performance QD display devices.
OPTICAL DESIGN AND TESTING XII
(2023)