MATLAB 影像處理及擴展應用研討會醫學影像 機器學習 物聯網 (IoT)

Transcription

1 MATLAB 影像處理及擴展應用研討會醫學影像 機器學習 物聯網 (IoT) Application Engineer Judy Yang

2 Agenda Time Topics Welcome 13:30-14:30 MATLAB 於影像處理介紹 14:30-15:20 應用機器學習技法於影像處理領域 15:20-15:30 Coffee Break 15:30-16:30 MATLAB 與物聯網 : 連結至穿戴式裝置硬體之訊號處理 Q&A and Wrap UP

3 Applications: Image and Video Processing Medical imaging Surveillance Robotics Automotive safety Consumer electronics Geospatial computing Machine vision and more

4 Different Aspects of Image Processing Importing, visualizing and investigating Enhancing images Aligning multiple images Segmenting images Detecting image features Classifying images

5 Technical Computing Workflow Access Explore & Discover Share Image data in files Visualization tools Data/Image Analysis and Modeling Reporting and Documentation Outputs for Design Image acquisition Image Processing Application Development Deployment Iteratively

6 Consider this image from the Centers for Disease Control: Our goal: To develop an algorithm to detect and quantify infection. How many cells are in the image, and how many are infected?

7 Quantifying infection across multiple images Despite widely varying image quality

8 Identify key challenges, consider strategies: Challenges: Differences in color Differences in illumination Contiguity of cells Low resolution/poor quality Strategies: Using apps to explore images Pre-processing Watershed segmentation Morphological segmentation DEMO

9 In this session we quantified rates of infection in heterogeneous images

10 What if we wanted to classify the type of infection, differentiating several species of parasites?

11 Agenda Time Topics Welcome 13:30-14:30 MATLAB 於影像處理介紹 14:30-15:20 應用機器學習技法於影像處理領域 15:20-15:30 Coffee Break 15:30-16:30 MATLAB 與物聯網 : 連結至穿戴式裝置硬體之訊號處理 Q&A and Wrap UP

12 Machine Learning A machine learning algorithm takes examples of inputs and outputs associated with a task and produces a program that can automatically differentiate them. Hand Written Program boats mugs hats Computer Vision Machine Learning boats mugs hats If brightness > 0.5 then hat If edge_density < 4 and major_axis > 5 then boat model = fitcsvm (image_features, label)

13 Machine Learning Workflow Using Images Training Data Feature Extraction, Encoding Machine Learning babesiosis plasmodium chagas Classifier babesiosis Input Image Feature Extraction, Encoding Classification

14 Bag of Words Image Processing Toolbox Class / Label Perform image processing, analysis, and algorithm development Image Processing Toolbox provides a comprehensive set of referencestandard algorithms, functions, and apps for image processing, analysis, visualization, and algorithm development. You can perform image analysis, image segmentation, image enhancement, noise reduction, geometric transformations, and image registration. Many toolbox functions support multicore processors, GPUs, and C-code generation. Image Processing Toolbox supports a diverse set of image types, including high dynamic range, gigapixel resolution, embedded ICC profile, and tomographic. Visualization functions and apps let you explore images and videos, examine a region of pixels, adjust color and contrast, create contours or histograms, and manipulate regions of interest (ROIs). The toolbox supports workflows for processing, displaying, and navigating large images. Training Data Bag: image processing, analysis, image, pixels, enhancement Vocabulary / Bag of Words

15 Bag of Visual Words ( features) babesiosis Class / Label Training Data Vocabulary / Bag of Features

16 Encoded images What is a Classifier? Training Data Features Classifier Machine Learning Classification babesiosis plasmodium chagas Class Membership

17 So let s give it a try DEMO

18 Using Machine Learning for Computer Vision Image Processing Toolbox Provides 100s of validated functions Indispensable for image processing applications Computer Vision System Toolbox Provides tools to generate image features for training classifiers See doc for full list of provided image features Statistics and Machine Learning Toolbox Provides learning algorithms to train classifiers

19 Agenda Time Topics Welcome 13:30-14:30 MATLAB 於影像處理介紹 14:30-15:20 應用機器學習技法於影像處理領域 15:20-15:30 Coffee Break 15:30-16:30 MATLAB 與物聯網 : 連結至穿戴式裝置硬體之訊號處理 Q&A and Wrap UP

20 The Challenge of IoT Devices? Insight

21 What is the Internet of Things? Integration Business Systems Data Aggregator (Server or Cloud) Deploy analytics to cloud Deploy algorithms to smart devices Smart Connected Devices Exploratory Analysis

22 Look at the Pieces: Devices Sensors and Human Interaction May have strict energy budget Required embedded programming skills Bandwidth is expensive in power and dollars Goal for device is to do as much data reduction locally as possible Smart, Connected Devices Communication Embedded Sensor Analytics Data Reduction

23 Look at the Pieces: Data Collection & Online Analytics Server or Cloud-based Log and analyze information across collection of devices Real-time information for situational awareness Requires cloud/web development skills and operations support Data Intake must be scalable and reliable True Big Data Data Aggregator Storage On-Line analytics Visualization & reporting

24 Look at the Pieces: Exploratory Analysis Desktop-based Access historical device information Analyze past performance for predictive modeling and deep insight Requires data analysis / data science skills Heavy use of statistics and signal processing techniques Deploy complex algorithms to both cloud and edge devices Exploratory Analysis Historical analytics Algorithm development

25 IoT Design challenges Business Systems Streaming data management and storage Online and real-time analytics Ability to visualize results and make decisions Integrating various software components Acquiring data from external sources Embedded development is challenging Increasing algorithmic complexity Need connectivity to cloud resources Limited power and data bandwidth Advanced analysis algorithms Gain insight into historical data Tools for deployment and connectivity

26 MathWorks Capabilities for IoT Leveraging MATLAB Analytics and Model-Based Design

27 Overview Integration Business Systems Model Device Communication Publish data to Cloud Data Aggregator (Server or Cloud) Data storage Online analytics Visualization and reporting Deploy analytics to cloud Deploy algorithms to smart devices Smart Connected Devices Embedded algorithms for sensing, data reduction, and control Models and simulation of system behavior Monitor and control nodes/devices Exploratory Analysis Historical analytics Algorithm development MathWorks Provides Capabilities for All of these Steps

28 MATLAB & Simulink Capabilities for IoT Deployment.NET, COM components Java components Multicore and GPU systems Spreadsheet plug-ins Database plug-ins Hadoop Cloud services (AWS) ThingSpeak Apps Smartphone/tablet integration File I/O Text Spreadsheet XML CDF/HDF Image Audio Video Geospatial Web content Real-Time Sources Sensors GPS Instrumentation Cameras Communication systems Machines: embedded systems fieldbus Financial datafeeds Repositories Databases (SQL) NoSQL Hadoop Communication Protocols CAN DDS OPC XCP Physical Component Modeling Electronic Mechanical Hydraulic, etc. Communications Protocol Modeling LTE, Zigbee, , etc. Automatic Code Generation Programmable chips (MCU, DSP, etc.) FPGAs Verification/Validation and Process Support Model- and Code proving Lifecycle management tools Data Clean-up Filtering Image processing Signal processing Telemetry RF sampling Analysis, Modeling, Design Data visualization Statistics Regression Machine learning (supervised &unsupervised) Neural networks Optimization (gradient-based & stochastic) Symbolic computing Image analysis Financial analysis Geospatial computing Object recognition Speech recognition

29 Why MATLAB? Engineering, Scientific, and Field Data Business and Transactional Data MATLAB and Simulink are well positioned for: A. Analytics that increasingly require both business and engineering data D A Analytics Design and Development B. Developing embedded systems which have increasing analytic content C. Deploying the increasing number of analytic-rich applications that run on both traditional IT and embedded platforms B Target to embedded systems C C Deploy and integrate in IT systems D. Enabling domain experts to do data science

30 Customer Case Study

31 Customer Study: isonea Cloud and Embedded Analytics Opportunity Develop an acoustic respiratory monitoring system for wheeze detection and asthma management Analytics in cloud and embedded Captures 30 seconds of windpipe sound and processes the data locally to clean up and reduce ambient noise Invokes spectral processing and pattern-detection analytics for wheeze detection on isonea server in the cloud Provides feedback to the patient on their smartphone Benefit Eliminates error-prone self-reporting and visits to the doctor

32 Sensor Analytics for IoT

33 Signal analysis for classification Application examples Mobile sensing Structural health monitoring (SHM) Fault and event detection Automated trading Radar post-processing Advanced surveillance...

34 Case Study: Human Activity Analysis and Classification Feature Extraction Classificatio n Dataset courtesy of: Davide Anguita, Alessandro Ghio, Luca Oneto, Xavier Parra and Jorge L. Reyes-Ortiz. Human Activity Recognition on Smartphones using a Multiclass Hardware-Friendly Support Vector Machine. International Workshop of Ambient Assisted Living (IWAAL 2012). Vitoria-Gasteiz, Spain. Dec

35 Sensor Data Analytics Workflow the bigger picture Signal Processing Machine Learning Domain knowledge Open-ended problem Long discovery cycles Built-in algorithms Concise code (54 lines for 66 features!) Apps and visualisation accelerate insight

36 Sensor Data Analytics Workflow the bigger picture Connect and Acquire Signal Processing Machine Learning Different tools and environments Disconnect between hardware and analysis Inefficiencies in data sharing MATLAB Connects to DAQ interfaces and sensors directly. E.g. Android Sensor Support iphone and ipad Sensor Support

37 Sensor Data Analytics Workflow the bigger picture Connect and Acquire Signal Processing Machine Learning Embedded Implementation Signal analysis vs. on-line DSP From Machine Learning theory to pre-trained, low-footprint classifiers MATLAB vs. C/C++ Streaming algorithms, data sources and visualization for System modelling and simulation Automatic code generation

38 Leverage Built-in Algorithms, Apps, and Technologies Signal Processing Toolbox Built-in algorithms and Apps to process and analyse signals cheby2 filter rms pwelch periodogra m xcov findpeaks

39 Leverage Built-in Algorithms, Apps, and Technologies Signal Processing Toolbox Parallel Computing Toolbox Accelerate computationally and dataintensive problems using multicore processors, GPUs and computer clusters parfor

40 Leverage Built-in Algorithms, Apps, and Technologies Signal Processing Toolbox Parallel Computing Toolbox >> classificationlearner Statistics and Machine Learning Toolbox Functions and apps to describe, analyze, and model data. Regression, clustering and classification algorithms to draw inferences from data and build predictive models

41 Leverage Built-in Algorithms, Apps and Technologies Signal Processing Toolbox Parallel Computing Toolbox >> nprtool patternnet Statistics and Machine Learning Toolbox Neural Network Toolbox Functions and apps to design, train, visualize, and simulate neural networks

42 Leverage Built-in Algorithms, Apps and Technologies Signal Processing Toolbox Parallel Computing Toolbox Statistics Toolbox BiquadFilter MatFileReader Autocorrelator SpectrumEstimator TimeScope Neural Network Toolbox DSP System Toolbox Streaming algorithms, data sources and visualization for system modelling and simulation

43 Leverage Built-in Algorithms, Apps and Technologies Signal Processing Toolbox >> codegen Parallel Computing Toolbox Statistics Toolbox Neural Network Toolbox DSP System Toolbox MATLAB Coder Generate embeddable source C/C++ from MATLAB code (Learn more: MATLAB to C Made Easy webinar)

44 Signal Processing and Machine Learning Techniques for Sensor Data Analytics Summary Extensive set of de-facto standard functions for signal processing and machine learning Environment accelerates insight and automation: visualisation, apps, language, documentation Path to embedded products, from on-line simulation to automatic code generation

45 Any Questions?