DEEP LEARNING APPROACH FOR REMOTE SENSING IMAGE ANALYSIS
DEEP LEARNINGAPPROACH FOR REMOTESENSING IMAGE ANALYSISAmina Ben Hamida*,**Alexandre Benoit*, Patrick Lambert*, Chokri Ben Amar*** LISTIC, Université Savoie Mont Blanc, France{amina.ben-hamida,alexandre.benoit, patrick.lambert}@univ-smb.fr** REGIM, ENIS, Tunisia, chokri.benamar@ieee.org
Presentation outlineScientific context Big Data Deep Learning (DL) Remote SensingDL for hyperspectral Data Experimental dataset DL architectures ResultsDiscussion & Future work
Scientific ContextSpecific Fields350 millions photosare uploaded dailyBig Data100 hours of videosare uploaded everyminute: 2 billionseach year1.4 millions of minutechats are saved everyminuteMedical Imaging .Remote Sensing:(RS)Use case example :Sentinel satellites whichprovide some thousands ofterabytes of data on a scaleof 10 years.
Scientific ContextCan we adapt recent methods developed in the multimedia community for RS ?
Deep LearningModelling high level abstractionsfrom multiple non linear transformations“Rachel”
Deep LearningFully connected layer : connects all theneurons to allavailable inputs No spatial embeddingNon linearity : Impact of convergencespeed !!!
Deep LearningConvolutional layer : Local filtering Rich feature mapsgenerationPooling layer : Subsamplingsignals Add translationrobustness
Hyperspectral Data
DL for Hyperspectral DataClassificationTaking into account the spatial and spectral componentsSeperatelyOnly using spectralEarly combining(using SAE)informationspatial and umberinformation more datafor training?Looks good
Experimental datasetUniversity of Pavia datasetSingle image610 340 pixels103 bands9 classes
DL architectureCascading 3D convolutions, 1D convolutions and final fully connected layers
Hyperspectral Deep Networkarchitectures3 layers3D/1D4 layers3D/1D6 layers3D/1D
Results :accuracy vs complexityAccuracy when training on 5% of the data10095.69593.893.9Accuracy9085651*14 layers010000200003000079.33*37570*5*56 layers7592.585.28486.680Spatialrangeimpact75.93 layers4000050000Number of parameters* Hu&al, “Deep convolutional neural networks for hyperspectral imageclassification,” in Journal of Sensors, 2015600007000080000
Results :accuracy vs complexityDeeper models for increased performances and lessparameters.Spatial information does matterbut spatial range depends on the use caseDeeper networks need more time to train
Results :6 layers deep net, 5*5 neighbors
Results :6 layers deep net, 5*5 neighborsSpectral profiles
Results :6 layers deep net, 5*5 neighborsPer class accuracy mostly stable 95% on averageClassification errors explained by : similar spectral profiles boundary effects (ROI size vs neighborhood class)
Results :confusion vs neighborhood3*31*15*5Processing time(caffe, CPU mode,Dual core i7 proc).1h2h5hObservation : spatial information gradually correctsspectral based errors
Results :Accuracy vs training dataset sizeAccuracyCNN challenger, 5*5neighbors, no pretrainingK. Makantasis&al “Deepsupervised learning forhyperspectral dataclassification throughconvolutional neuralnetworks,” IGRS2015 20000 parametersAccuracy on Pavia University dataset1021009896946 layers, 3*3 neighbors, 4419 parameters6 layers, 5*5 neighbors, 6074 parameters9290880102030405060Training samples ratio (%)708090100SAE challenger, 7*7neighbors, with pretrainingX. Ma&al“Hyperspectral image classification viacontextual deep learning,”EURASIP JIVP 2015 20000 parameters
ConclusionDeep Learning can do the job ! Automatic adaptation to the context and good results Deeper is better. up to a limit ?Main issues : Expertise required Network architecture design Training procedures design Reduce the number of parameters
Future Work guidelineEnhance architecturesSiameseLearning metrics from similarity measuresNetworksSqueezeNetGet lighter models !approachAdapt to new contextsTheSwitch to multispectral dataSentinelUse casePlay with unlabelled data35
What's next ?Yes, DL was so far so good for simple RS applicationBut, what gaps will it be facing when hardening thetask ?Questions ?32
Thank you for your attention
Results :from one dataset to anotherAccuracy vsdataset, deepness,neighborhood3 layers6 layersPaviaUniversityPaviaCenter1*1 neighbors75.9 %90.5%3*3 neighbors84.0 %94.5%5*5 neighbors93.8 %96.4%7*7 neighbors85.9 %96.2%1*1 neighbors86.5 %3*3 neighbors92.3 %5*5 neighbors93.8 %98.5%
Future Work guidelineTesting the robustness level of the DL structureFacingInjecting noise into the system in order to test its ability to dealNoisewith noisy images.DegradeTesting to what extent can the system face a variety of trials toperformancesdegrade its performances.37
Future Work guidelineRelying on larger ground truth databasesLargerThe use of other dabases in order to create ground truthamount ofannotaded ones.dataThis work can be done in collaboration with other labs.38
Future Work guidelineExtending the work to the sentinel databasesResorting to multispectral and hyperspectral data, withThecomplex challenges to rise.SentinelUse caseFacing the challenge of large unlabelled data40
Conv layer hintsparameters vs IO dimensionsmi nfli f24
APPROACH FOR REMOTE SENSING IMAGE ANALYSIS * LISTIC, Université Savoie Mont Blanc, France {amina.ben-hamida,alexandre.benoit, patrick.lambert}@univ-smb.fr ** REGIM, ENIS, Tunisia, chokri.benamar@ieee.org Amina Ben Hamida*,** Alexandre Benoit*, Patrick Lambert*, Chokri Ben Amar** Presentation outline Scientific context Big Data Deep Learning (DL) Remote Sensing DL for hyperspectral Data .
Bruksanvisning för bilstereo . Bruksanvisning for bilstereo . Instrukcja obsługi samochodowego odtwarzacza stereo . Operating Instructions for Car Stereo . 610-104 . SV . Bruksanvisning i original
10 tips och tricks för att lyckas med ert sap-projekt 20 SAPSANYTT 2/2015 De flesta projektledare känner säkert till Cobb’s paradox. Martin Cobb verkade som CIO för sekretariatet för Treasury Board of Canada 1995 då han ställde frågan
service i Norge och Finland drivs inom ramen för ett enskilt företag (NRK. 1 och Yleisradio), fin ns det i Sverige tre: Ett för tv (Sveriges Television , SVT ), ett för radio (Sveriges Radio , SR ) och ett för utbildnings program (Sveriges Utbildningsradio, UR, vilket till följd av sin begränsade storlek inte återfinns bland de 25 största
Hotell För hotell anges de tre klasserna A/B, C och D. Det betyder att den "normala" standarden C är acceptabel men att motiven för en högre standard är starka. Ljudklass C motsvarar de tidigare normkraven för hotell, ljudklass A/B motsvarar kraven för moderna hotell med hög standard och ljudklass D kan användas vid
LÄS NOGGRANT FÖLJANDE VILLKOR FÖR APPLE DEVELOPER PROGRAM LICENCE . Apple Developer Program License Agreement Syfte Du vill använda Apple-mjukvara (enligt definitionen nedan) för att utveckla en eller flera Applikationer (enligt definitionen nedan) för Apple-märkta produkter. . Applikationer som utvecklas för iOS-produkter, Apple .
Deep Learning: Top 7 Ways to Get Started with MATLAB Deep Learning with MATLAB: Quick-Start Videos Start Deep Learning Faster Using Transfer Learning Transfer Learning Using AlexNet Introduction to Convolutional Neural Networks Create a Simple Deep Learning Network for Classification Deep Learning for Computer Vision with MATLAB
Table of Contents Purpose of this Document 5 Our Approach to Remote Learning 6 Equity as Our Core Identity 7 What We Learned in the Spring 8 What We're Doing Differently 9 The Remote Learning Plan 10 Best Practices for Remote Learning 11 Fluidity Across Scenarios 12 Remote Learning 13 Fully Virtual/100% Remote Program 14 Learning Management System and Curriculum 15
2.3 Deep Reinforcement Learning: Deep Q-Network 7 that the output computed is consistent with the training labels in the training set for a given image. [1] 2.3 Deep Reinforcement Learning: Deep Q-Network Deep Reinforcement Learning are implementations of Reinforcement Learning methods that use Deep Neural Networks to calculate the optimal policy.
