“复杂环境光电信息感知科学与技术”国际学术交流平台2021年学术年会

2021 International Workshop on Optoelectronic Perception (IWOP2021)

Oct.15th-18th, 2021, in Xi'An, China

第五届“复杂环境光电信息感知科学与技术国际学术交流平台”学术交流年会定于10月15-18日在线上/线下同步举行。本次会议主要围绕复杂环境光电信息感知领域的相关问题展开讨论，将邀请国内外权威专家学者就其最新研究成果及前沿发展等作相关报告，旨在为本领域科技工作者，包括海外留学人员，提供一个展示技术创新，推动技术应用，探讨合作的开放学术平台。本次会议无注册费用，投稿论文全文收入年会内部文集。

一、会议主题

关于复杂环境与目标光电特性、电磁散射特性及电波传播特性、光电探测技术、光电信息与图像处理、遥感图像处理、计算成像、生物医学成像等领域的基础理论、关键技术和应用研发的相关科学问题。

二、重要日期

会议日期：2021年10月15-18日

摘要投稿截止日期：2021年8月31日

全文投稿截止日期：2021年9月30日

三、会议组织

主办单位：西安电子科技大学物理与光电工程学院

协办单位：西安电子科技大学国际合作与交流部

四、会议地点

线上/线下（西安）会议

五、会议邀请报告

(1). Prof. Chein-I Chang：IEEE Life Fellow, Fellow of SPIE. Professor with Department of Computer Science and Electrical Engineering at the University of Maryland, USA

Talk Title: Hyperspectral Anomaly Detection and Its Performance Evaluation by 3D ROC Analysis

Abstract: Hyperspectral anomaly detection has received considerable interests recently. Many approaches have been proposed and developed and its performance has been evaluated by the traditional 2D receiver operating characteristic (ROC) analysis where the area under an ROC curve plotted based on detection probability, PD versus false alarm probability, PF, AUC(D,F). Unfortunately, many ROC curves reported in the literature are indeed incorrectly generated. Another major issue is that solely relying on AUC(D,F) is unreliable and misleading as will be demonstrated in this talk. This is because PD and PF are generated by the same threshold. As a result, a higher PD also generates a higher PF and vice versa. In particular, since anomaly detection does not have any prior knowledge, background suppression is more crucial and important than target detectability. Accordingly, using AUC(D,F) alone for anomaly detection does not address the background suppression issue. To address these two issues this talk presents a 3D ROC analysis which generates a 3D ROC curve as a function of (PD,PF,) by  including  the threshold parameter  as a third independent variable. Consequently, a 3D ROC curve along with its derived three 2D ROC curves of (PD,PF), (PD,) and (PF,) can be further used to design new quantitative detection measures to evaluate the effectiveness of a detector and its target detectability TD and background suppressibility (BS). To demonstrate the full utility of 3D ROC analysis in target detection, examples are included in this talk to demonstrate how 3D ROC curves can be used to design new detection measures to evaluate target/anomaly detection performance more effectively ad accurately in terms, TD, BS and detector’s effectiveness.

        (2). A. Prof. Xiuping Jia, IEEE Fellow,  The University of New South Wales at Canberra, Australia

Talk Title: Information Mining at Multiple Scales from Remote Sensing Digital Images

Abstract:  Remote sensing digital images are widely available nowadays for a wide range of applications, including landcover and land use mapping, target detection, urbanization and climate change monitoring. Machine learning or deep learning is a critical tool to achieve information mining from the raw image data, such as optical multispectral and hyperspectral data. It however faces a challenging need for a multiscale data analysis to accommodate the nature of individual problems. Image analysis is also responsible for compensating limited data quality in terms of spatial resolution and signal-to-noise ratio to improve information mining reliability. As a result, subpixel and superpixel analysis has been developed. The traditional pixel classification is achieved by incorporating spatial information and making use of the spatial correlation property. Scene (patch, image) classification has become easier with the greatly increased computing power and the development of deep Neural networks in recent years.
In this talk, classification at various scales will be presented and overviewed. In supervised learning, modeling and training are the fundamental steps. Overfitting is a big issue to address, which will be discussed in this talk.

(3). Prof. Artem Pliss, University at Buffalo, USA

Talk Title: Emerging optical approaches for quantitative cell biology research

Abstract:  Research in fundamental cell biology and pathology could be revolutionized by developing the capacity for quantitative molecular analysis of live cell organelles and other subcellular structures. With this in mind, here will be presented recently developed directions in biosensing, aimed at probing of physico-chemical environment in either live or fixed cells. First, we introduce the application of fluorescence lifetime imaging (FLIM) for mapping the density of macromolecules in subcellular structures. This assay utilizes a correlation between the fluorescence lifetime of fluorophore and the refractive index of its microenvironment varying due to changes in the concentrations of macromolecules, mainly proteins and nucleic acids. Another FLIM approach, presented in this talk, was developed for sensing of the DNA compaction; it involves the process of Förster resonance energy transfer (FRET) between the donor and the acceptor fluorophores, both incorporated into DNA. Next, will be presented recent advancements in the Raman spectrometry, which at the current stage of development gives a rise to a Ramanomics - an independent optical Omics discipline. This emerging Omics approach categorizes the entire molecular makeup of a sample into about a dozen of general classes and subclasses of biomolecules and quantifies their amounts in submicron volumes, such as single organelles. A major contribution of this direction is an attempt to bridge Raman spectrometry with big-data analysis in order to identify complex patterns of biomolecules in the tissues and leverage discovery of disease biomarkers.

(4).A. Prof. Zhengbao Yang, The City University of Hong Kong

Talk Title: Piezoelectric Materials for Energy harvesters and flexible sensors

Abstract:  With the rapid advances in wireless sensors, implantable electronics, and wearable devices, the demand for high power-density and long-lifespan power sources is becoming increasingly stronger. Energy harvesting, emerging as an alternative energy solution to batteries, holds great potential to achieve self-powered autonomous operations of such low-power electronic devices, and thus has recently attracted much attention from both academia and
industry. The piezoelectric effect is widely adopted to convert mechanical energy to electrical energy, due to its high energy conversion efficiency, ease of implementation, and miniaturization. Piezoelectric effect, since its first description by the Curie brother in 1880, has become one of the most important effect used in the electronic industry. In the the Smart Transducers and Vibration Laboratory (STVL), we conduct research on different fabrication techniques for piezoceramics PZT, BTO, lead-free BCZT and piezopolymer PVDF in both MEMS and human scales. In today’s talk I will introduce our recent research on vibration energy harvesting and human tactile sensors using piezoelectric materials.

(5). Dr. Wei Lin, University of Technology Sydney, Australia

Talk Title: Omnidirectional Vertically, Horizontally, and Circularly Polarized Antennas and Array

Abstract:  Omnidirectional antennas have 360˚ coverage in the azimuth plane and thus are desired for point-to-multipoint and the emerging device-to-device (D2D) wireless communications. A high directivity, compact, omnidirectional horizontally polarized (OHP) antenna array is developed for wirelessly powering internet-of-things (IoT) devices. The antenna array is realized by seamlessly inserting several phase inverters inside an electrically long TE0.5,0 mode open waveguide. The phase inverter consists of a meandered slot and eight shorting vias. The meandered slot creates an interdigitated structure on the top surface of the waveguide; it introduces capacitance. The eight shorting vias are placed in an alternating pattern on the two sides of the slot; they produce inductance. The combination of the slot and vias forms a bandpass effect and inverts the electric fields in the waveguide. Consequently, a collinear and in-phase magnetic dipole array is realized. A compact eight-element OHP magnetic dipole array is designed, fabricated and measured. The measured results confirm the design concept and high directivity (10.4 dBi), omnidirectional HP radiation pattern has been achieved. Moreover, the most recently developed omnidirectional vertically-polarized and circularly-polarized antennas and arrays by Dr Lin and Prof. Richard W Ziolkowski at University of Technology Sydney will also be discussed in this talk.

         (6). Prof. Steven Gao : IEEE Fellow, University of Kent, UK

         Talk Title: Broadband Dual-Polarized Multi-Beam Millimeter-Wave and Sub-THz Antennas for Joint Communications and Radar Sensing Systems

         Abstract:  Both the Mobile Communications and the Radar Systems have experienced fast development during the past four decades. Recently it has become increasingly important for the seamless convergence of communication and radar so as to form a Joint Communications and Radar Sensing (JCRS) System and Network. Such a JCRS system and network turns the mobile communication network into a ubiquitous radar perception system, thus it is promising for many applications such as autonomous driving, lower altitude air traffic control, etc.  This talk will present several recent examples of broadband dual-polarized multi-beam antennas at millimeter-wave and sub-THz frequency bands for future JCRS systems and networks. This talk has two parts. First, an overview of recent development in dual-polarized antenna arrays at mm-wave and sub-THz frequencies will be provided, and different dual-polarized antenna technologies will be discussed. Then, several recent examples of mm-wave and sub-THz antenna arrays, developed at the University of Kent, UK, together with the collaboration partners, will be presented. These include one broadband dual-polarized slanted ±45° millimeter-wave array antenna having low cross-polarization levels, high isolation and high aperture efficiency, one millimeter-wave dual-polarized multi-beam array antenna with two-dimensional beam switching, one scalable dual-polarized millimeter-wave array antenna with low complexity and high cross-polarization discrimination, and one multi-beam array antenna at sub-THz frequency band. The results will be discussed. During the talk, the designs of these four array antennas will also be explained. Finally, a discussion of future challenges will be given at the end.

 (7). Dr. Guangwei Yang，Queen Mary University of London, UK

 Talk Title: Shared-Aperture Technology for Beam-Scanning Antennas

 Abstract:  In the last decade, with the development of wireless communication technologies such as the internet of things (IOT) and intelligent transportation systems (ITS), satellite communications, higher requirements have been placed on the high performance, compact, multi-function, and reliability of antenna systems. Shared-aperture technology is a very good solution and has been developed in various ways, such as hollowing, nesting, overlapping, etc. However, as higher design requirements are introduced, shared-aperture techniques need to keep pace with it and this presentation will introduce a new shared-aperture technique that can be applied to antenna systems.

(8). Prof. KuanFang Ren, First class professor in Rouen University, France

        Talk Title: Ray theory of waves for the scattering by large non-spherical objects

        Abstract: The interaction of waves (light, electromagnetic or acoustic waves) with macroscopic objects concern great number of research domains. Though the fundamental laws (Maxwell equations, Euler equation, etc.) have been well established, their applications to different physical phenomena or in engineering are always challenging. For example, we have not efficient methods to predict with precision the scattered field of a wave by large objects of complex forms, such as radar signal of an aircraft or scattering pattern of an oblate spheroid or a raindrop.

(9). Prof. Fabrice R.A. Onofri，Aix-Marseille University, France

Talk Title: Experimental analysis and modelling of spray drying of colloidal suspensions

Abstract:  Spray drying technology is commonly used in numerous process industries. It is extensively used for food processing, and in polymer, pharmaceutical, and porous material engineering, to obtain powders composed of solid particles with well-defined characteristics, such as particle morphology, size distribution, porosity, and density. Among the many droplet-to-particle drying patterns, several typical behaviours can be distinguished. This talk will deal with the case where droplets show an agglomerative tendency. In this case of importance (i.e. colloidal suspensions), the solid nanoparticles, suspended in an aqueous solvent, are sprayed with an assisted nozzle into a chamber, which is supplied with an additional hot gas flow. The droplets are transported by the gas, while getting dried. The nanoparticle agglomerates (referred as “particles”) essentially collect at the bottom of the spray chamber when the solvent within the droplets is fully vaporized. The control of both unit operability and outlet particle properties requires a thorough understanding of various complex phenomena, which include: (i) interaction of the gas and solid flows in the chamber; (ii) drying mechanism at the particle scale; (iii) particle–particle collisions; and (iv) particle adhesion on the walls. Here, we briefly review and understand how these phenomena are fundamental. The modeling of particle transport in a gas flow requires extensive computational fluid dynamics (CFD) calculations, which are based on either the Euler–Euler or Euler–Lagrangian methods. However, despite the recent developments in the CFD modeling of spray dryers, the operating conditions required to obtain powders with well specified properties, such as morphology and specific area, are still essentially established through an empirical approach. Such an approach is not only computationally expensive and time consuming, but also often inadequate for scaling up the process systems and pilot plants, and for extensive parametric studies. Thus, any affordable and efficient modelling approach that takes into account the basic particle drying mechanisms and the hydrodynamics in the spray chamber is highly desirable.
With this perspective, this talk will report the work in progress to develop an efficient numerical model that accounts for the drying process at the droplet scale, and the interactions and transport phenomena at the scale of the spray drier unit. The numerical results and discussions are supported by experiments carried out on boehmite and silica colloidal suspensions using a mini spray drier and an evaporating chamber equipped with an integrated acoustic trap and optical diagnostics (shadowgraphy and rainbow diffractometry). After introducing the background of this study, this talk will be focused on the key features and operating modes of the two experimental setups, i.e., an acoustic trapping experiment and a mini spray dryer. Then the coupled models referred to as the droplet drying and spray drying models will be summarized. Afterwards, experimental and numerical results will be presented and discussed, prior to conclude on actual achievements and perspectives.

(10). A. Prof. Jun Zhou. Griffith University, Australia.

Talk Title: Relative Depth Estimation from Monocular Hyperspectral Image

Abstract:  Relative depth estimation from images is an active research topic in computer vision. Traditional methods use grayscale or color images as input, but seldom study spectral properties for this task. This talk introduces the first work on depth estimation from hyperspectral images by exploring both intrinsic and extrinsic properties of the hyperspectral images. We propose that change in focus across band images of hyperspectral images due to chromatic aberration and band-wise defocus blur can be integrated for depth estimation. Novel methods are developed to estimate sparse depth maps based on different integration models. By adopting manifold learning, an effective objective function is developed to combine all sparse depth maps into a final optimized sparse depth map. Lastly, a new dense depth map generation approach is proposed, which extrapolates sparse depth cues by using material-based properties on graph Laplacian. Experimental results show that our methods successfully exploit spectral-spatial properties to generate depth cues.

(11). Prof. David Suter,  Edith Cowan University (Perth, Western Australia)

Talk Title: If (deep) learning is the methodology of choice – what do we miss?

Abstract:  Deep learning is the dominant “paradigm” now for computer vision. The original successes were in problems/areas lacking in good mathematical models (what is a cat? What is the structure of an image of a cat?); but has now impinged heavily even in areas where we thought we had good models (e.g., structure from motion). Of course, a paradigm doesn’t become dominant and successful, at least in an applied or engineering area, unless it successfully produces results. Moreover, if deep learning (and other learning based) approaches continue to deliver – should we seek anything more? What do we miss if we do everything by deep learning? Well, for one thing, we don’t produce a thorough (some might say, meaningful, insightful) understanding of the problem we are tackling. To hazard a sweeping generalisation: one might say that deep learning is characterised by “plumbing” (selecting successful modules for which we only have a vague characterisation of “what they do, and how they do it”), and dreaming up ways to tweak these modules or to interconnect these in useful ways. Gone is the careful problem definition, and the analysis of that problem (and how it might relate to other problems or a useful general framework).
On the other hand, “traditional model based” computer vision at least affords the opportunity to discover surprising connections between problems. I will illustrate this for one area I know reasonably well: robust fitting – but more specifically, robust fitting using the maximum consensus criterion, a.k.a, MaxCon. This is a problem that has been studied in computer vision for 40 years; and yet, only recently, has been demonstrated to connect to a significant body of mathematical theory/structure. It is also linked in surprising ways to some standard “prototypical” computer science problems.

(12). Dr. Lin GU, RIKEN AIP, Japan

Talk Title: Beyond RGB: Transfer Image Knowledge to Multimodal Learning

Abstract:  The deep learning has achieved much success on RGB images. This talk will discuss how to transfer the knowledge learned from standard vision to the real-world task on multi-modal data including medical imaging, low-light vision, hyperspectral imaging and etc.The deep learning has achieved much success on standard RGB images taken under clear environment. Here, we are working on dealing with real-world challenges where the data may come from multiple sources. For example, the medical image analysis needs to process the MRI, CT images which is quite different from normal RGB images. In addition, incorporating multiple-modal data such as the diagnose report and expert gaze information has proven promising enhancing the performance existing medical research.
In this talk, we will show how to utilise the knowledge of standard vision, including the architecture and pre-trained parameters, to extent to multi-modal learning on medical imaging, low-light vision, hyperspectral imaging and etc.

(15). Prof. Martin P. J. Lavery, University of Glasgow

Talk Title: Machine-learning-enhanced spatial spectroscopy for environmental sensing

Abstract: In this presentation we will outline our recent results utilising support vector network machine learning approaches to determine independently variations in Reynolds number and Fried Parameter over an atmospheric channel by analysing optical degradations in optical beams carrying Orbital Angular Momentum (OAM). Through numerical modelling of cascaded optical perturbations, a comprehensive training set of OAM mode spatial spectra was produced over a simulated 1.5km's free-space optical channel in an urban environment. Our results indicate this machine learning approach will determine independently the Reynolds number and Fried Parameter with over 90.4% accuracy. These results indicate potential new methods for determination of variation in material properties that could be used for the detection of environmental contamination and weather monitoring.

        (16). Prof. Zeev Zalevsky, IEEE Fellow, Bar-Ilan University, Israel.

Talk Title: Remote photonic sensing of hemodynamic brain activity

Abstract:  We will present a technological platform that can be used for remote sensing of hemodynamic activity in the brain. The technology is based upon illuminating the head with a laser and then using a properly defocused imaging optics connected to camera. The captured set of time changing secondary speckle patterns are temporally and spatially analyzed using special artificial intelligence (AI) algorithms in order to extract nano vibrations occurring in the illuminated volume. Those vibrations or movements are associated with hemodynamic processes occurring in the head when different cortexes are stimulated.This sensing concept was already applied for remote and continuous estimation of various vital bio signs such as heart beats and their variability, respiration, blood pulse pressure, intra ocular pressure and more. It was used for hematological applications such as for remote sensing of alcohol and glucose concentrations in blood stream and recently was applied for remote sensing of hemodynamic activity in the body.When the sensor is directed towards the head, we can sense from a distance the stimulation of various cortexes such as the visual and the motor cortex, cortexes related to different senses such as hearing, smell and taste senses as well as other types of cortexes such as the short memory cortex. Results related to applying this photonic sensor in biomedical experiments will be presented and discussed.

(17). Prof. Zhen Yuan, University of Macau

Talk Title: Temperature-feedback Nanoplatform for NIR-II Penta-modal Imaging-guided Synergistic Photothermal Therapy and CAR-NK immunotherapy of Lung Cancer

Abstract:  In this study, to visually acquire all-round structural and functional information of lung tumor while performing synergistic photothermal therapy (PTT) and targeting immunotherapy, a theranostic nanoplatform that introduced upconversion nanoparticles (UCNPs) and IR-1048 dye into the lipid-aptamer nanostructrure (UCILA) was constructed. Interestingly, the IR-1048 dye grafted into the lipid bilayer can serve as the theranostic agent for photoacoustic imaging, optical coherence tomography angiography, photothermal imaging and PTT in the second near infrared (NIR) window. In addition, loaded in the inner part of UCILA, UCNPs possess the superior luminescence property and high X-ray attenuation coefficient, which can act as contrast agents for CT and thermo-sensitive up-conversion luminescence (UCL) imaging, enabling real-time tracking of metabolic activity of tumor and temperature-feedback PTT. Further, under the complementary guidance of penta-modal imaging and an accurate control of in-situ temperature change during PTT, UCILA exhibited its excellent capability for ablating the lung tumor with minimal side effects. Meanwhile, synergistic CAR-NK immunotherapy was carried out specifically to eradicate any possible residual tumor cells after PTT. Therefore, the UCILA nanoplatform was demonstrated as a multifunctional theranostic agent for both penta-modal imaging and temperature-feedback PTT while conducting targeting immunotherapy of lung tumor.

(18). Prof. Ajmal Saeed Mian, The University of Western Australia.

Talk Title: Deep Learning over 3D Point Clouds

Abstract:  3D point clouds are becoming an important data source for vision tasks such as autonomous driving and robotic perception. However, deep learning over unstructured point clouds is challenging. We propose a spherical convolution kernel combined with octree guided neural network architecture for deep learning from unstructured point clouds. Spherical kernels systematically quantize point neighborhoods to identify local geometric structures and avert dynamic kernel generation during network training for efficiency. We incorporate fuzzy mechanism into our discrete spherical convolutional kernel to avoid boundary effects during learning and varying point density during inference. We also propose an efficient graph convolutional network SegGCN that exploits ResNet like encoder blocks and 1x1 convolutions in the decoder. Next, we release Picasso, a CUDA-based library for deep learning over complex real-world 3D meshes. Picasso contains CUDA implementations of our Spherical Kernel, Fuzzy Spherical Kernel, three additional convolution kernels (vertex2facet, facet2vertex, facet2facet) designed for 3D meshes and various network architecture design tools such as pooling, unpooling, residual layers, separable convolution etc. Picasso also contains the first CUDA-based on-the-fly mesh simplification algorithm to facilitate hierarchical deep learning. We show the effectiveness of our novel convolution kernels, network architectures and mesh simplification algorithm on synthetic and real-world datasets. Finally, I will discuss our data capture hardware, challenges of outdoor LiDAR data and our open source annotation tool that can be used to generate large scale datasets for deep learning. Our methods are published in CVPR and PAMI.

持续更新中

六、会议注册

免费注册。