thank you for introduction basic uh my name is wendy no postdoc from ghent university
that they i'm gonna put in that use something about this classification of hyperspectral image
oh would happen easily based on extended morphological profiles responsive reconstructions
different from the at normal actually image hyperspectral sensor cup two dates bits hundred append
but it's pixels
hot very hot it's hot continuing with reflections different materials have different spectral reflections
and we can identify different materials according to its reflections for instance here
according to this spectrum reflects the reflections we notice these soil and the this reflections
basis what the
for some hyperspectral image
the spectral reflections of some of pets are similar when only using spectral information to
do classification we cannot get sucked is final result spatial information it's very important spatial
information such as object size shape and texture it's very important to improve the classification
result
when they called by
both spectral information and the spatial information we can improve the classification accuracy a lot
this is all light of my presentation today first i will give a lot of
you about feature extractions at the morphological profiles which are widely used to explore ohms
the spectral and spatial information of hyperspectral image then i mean in that use all
propose it can be morphological profiles we sparse reconstructions and show some experimental results of
the last but not least it conclusion and future work
hyperspectral image it's difficult to process due to its high dimensionality
feature extraction
it's use it's always used as preprocessing to reduce the dimensionality and the lead then
there's the of the hyperspectral states
uh the
pretty simple components and the cannot pretty simple can note component devices are most widely
my set used to do basic directions
principal component analysis extracted features on which the tomb melody alliance
it's not seem to maximize
the first few extracted features can you percent most of a information of heidi by
original hyperspectral states and the last are mainly noise for instance here
the first we extracted pretty simple components can be presented ninety nine percent of Q
melody but lines of this one hundred and three bands hyperspectral target
and the so as the identity i dependence component and
to achieve the same ninety nine percent
Q melody lines can opening simple component needs poles
twelve can uh features extracted features but this as a nonlinear mess that's kernel principal
component the licensees can proceed that non linear distribution of that eight
and uh you see even on the total extracted features it's have some detail features
but for pretty simple and the linear ica in the in its twelve extracted features
maybe the noise
uh spatial information it's important especially for the high low solution up and hyperspectral image
in particular morphological profiles it's one of the most popular myself which are used to
explore the spatial information of hyperspectral image
that's take morphological closing as example to see how morphological profiles pretty C the spatial
information of hyperspectral date
in morphological closings a structure it demands are moved in inside the image when there's
the size of the dark object it's more that then these structural elements
then the object will be removed
at the size of this structure demands increase more and more dark
small object uh disappear in the disappear in this image
and then the size of a small object only like it this the size of
the strict demons
so we say that's morphological profiles contain yeah yes thank you
uh yeah thank you thank you much yeah
and the this size it's like you like it to the size of the structure
you meant
and the when the size of the structure immense increase
more and more objects more dark objects disappear we can see also the shape of
objects also chant
for the corn the police like tango
column is now a change to be lumped
well um
okay i was despising
that's take
one early example on how the a hyperspectral image to see the results when the
size of structure demands increase we see a lot of
bright
small objects disappear i don't know what you can see these uh small disappear uh
this one this one this one and the a small bright small objects disappear us
the size of structuring elements increase
but we can also see that the shape of this object at change
you see this lack tango object now the corner it's long it long
in order to proceed the shape of objects while always use morphological profiles these reconstructions
this christmas that the all the site the shape of objects can be preceded by
the then this know the constructions
we see here this object
the each of these black tango object it's we'll proceed how well the we see
that some small object which are expected to disappear now
state present in the image for this small loads point load by load which are
disappear in here we solve the constructions but now disappear we can say that this
reconstructions the size of object mott proceeds about the remotes proceed the well
in order to overcome all these problems morphological profiles this posture become structures are used
this is not set we can see us the size of the structure it immense
increase
more and more small point
objects disappear corresponding to i
structuring elements size at the same time we can proceed the age of the object
button then result the constructions
so we propose
uh extended morphological profiles this posture become could decode reconstructions to combine spectral and the
spatial information of the hyperspectral image
this is our proposed by set we take features extracted by pca as example to
build extended morphological profiles responsive reconstructions we first use the feature extractions delete use that
dimensionality and the led dependency of the hyperspectral data and then repute we use all
morphological openings and closings this partial reconstruction is to build morphological profiles on each extracted
features
and then we stack all these features as well
as important as the svm classifier and then finally we got the classification maps
that's take one example take two extracted features as pca to see we take two
we take two openings and the closings and the we start these features us
uh so but uh this is extended morphological profiles and the we use this as
input to do the classification
we can see by increasing the size of the structure immense more and more objects
are removed
and the extended morphological profiles county the spatial information such as the size of objects
in the image
the morphological profiles contain
yeah O explained this morphological profiles contain four openings and the full closings
and the this morphological profile with partial reconstruction is was proposed in to solve i
it's with peak
last row some experimental results to see if wishes all of all proposed method
the four states we use it's problem
university of probably i italian the states have one hundred and the fifteen bands
some bands are removed due to the noise leaving one hundred and sleep and on
the left this is the false color image it's
this image has high spatial resolution is one point three meters topic cells the online
classes in this image
we select ten labels impose a class landlady to change svm classifier based
a pf kernels
and then test on the testing on the test set
morphological features up builds on the first poll can open a simple components
and the lease is the results we used original hyperspectral image to do the classification
and this results it's
we use that we the morphological profile with no reconstruction is and this is the
morphological profile with reconstruction and this is all proposed which posture because structures
phone to be zero maps we can see that's this more formal logical features the
classification maps it's much smoother than results what spatial information
this is noise
and the compared to the morphological risk we construct use morphological profiles result reconstructions and
vese posture because actors a much and smooth then this one
and the full morphological profiles with no reconstruction is the shape of some object a
chant
for this model the corner here it's no longer
and we can also compare the overall classification accuracy here
here this the morphological features the overall accuracy can improve a lot even then what
kind that sense and that all proposed mess that what's the best results
and the beacon please see the size and shape but the then some existing methods
let's see some results when the change simple size increase
the extended morphological profiles built on different features because C
the morphological features proposed you on the
pretty simple component features
as the simple size increase all proposed it can be morphological profiles this posture reconstructions
got busted results
and
the morphological profiles you on kernel pretty simple components and lies is always get the
possibly start
in some cases it is button want to use the morphological profiles with reconstructions for
here this morphological profile
we speak objections cannot will be will be was then this no reconstructions this is
due to the old reconstructions of morphological profiles by reconstruction
let's see and not the plates that's a list that's it's prone city centre a
party a
it's has one hundred fifteen bands and some bad a limited due to the noise
leaving one hundred and two bands on the left side it's the false color image
is it also have high spatial resolution is one point three meters but excels yeah
nine different classes in this image
we select pen labels impose a class blunder the problem this training set
and the land passed on the touch screen set of this
morphological features appear on the top the first can cut kernel principal components
and the this is the result from people and that's that for a while what
dates we just use only the original hyperspectral data to do the classification and this
is no reconstruction this is this reconstruction and this is with all proposed
posture reconstructions because shape on that a little overall classification accuracy all postmaster what's the
highest accuracy
the see the performance when the change simple size increases
but the changes impose size increase all propose
extended morphological profiles with partial reconstructions got busted results and that this is always you
on the kernel principal components
so that's see the conclusion and the future work
uh in this presentation we can see that's kernel pretty simple component and vice is
more efficient to extracted features for comes construct in uh extended morphological profiles and the
extended morphological profiles with partial reconstruction is more efficient competing T then lowest be snow
constructions and the miss cost reconstructions
and the extended morphological profiles responsive reconstructions europe on kernel pretty simple components are always
perform fast especially before there's more training symbols
all future work maybe explore different structure meant structure element because now in this presentation
and the in my complex paper we just use the circle structure element and that
we can see also this kernel principal component we need more features
so it is but that that's it's great eight high dimensionality of the extended morphological
profiles this posture reconstructions
uh that's my presentation thank you
yeah
uh yes uh i prepared to try the linear structure elements which may efficient for
the loans for the streets and the label of the streets they are long lines
and the for the labels also like this one i think it's uh maybe it's
nice feature to improve the classification accuracy
yes
a yes uh yes we can use some computers before the classification this may have
some i think maybe some the same result because you smoother the original hyperspectral image
and the this different scalars you also can smooth the image
in different
uh how this yeah maybe can see that that's what if devoicing possibly denoising the
hyperspectral image and that we use that you noise image to do the classification
yes
so secure for suggestion i also read some level as follow your group yes uh
the extent it's attribute profiles yes it's but uh and the is we because see
that because uh it's sometimes it not related with this structure you meant so in
may be our future work we can see the we use the attributes uh profiles
to improve the classification accuracy that's good for test thank you
uh
yeah uh we normally we selected this according to the queue mel other human let
steve
cumulative but like the lions
for pca for kernel principal components the first we extracted features three percent ninety nine
percent of the two melodies alliance so for kind of pca how the for stole
components a reasonable kernel principal components people are sent ninety nine
a sense of the cumulative alliance so we just depend on the
yeah i don't different dates that's we have different
the different number of the extracted features we use yeah
yeah
yeah that's this also creates a new challenge because you also the dimensionality of the
extended morphological profiles will be higher than other than principal component like this
i