search

chameerawijebandara / Satellite-Image-Analyzer-

0 stars 0 forks source link

Finalize the NN structure #3

Open chameerawijebandara opened 9 years ago

chameerawijebandara commented 9 years ago

Finalize the good neural network stuser to implement

chameerawijebandara commented 9 years ago

@HBMDDS Did you Design a NN?

HBMDDS commented 9 years ago

I implemented the Neural Network code to recognize hand writings of '1' and '0' using some sample images of hand writing in internet. I did it to confirm whether my code is working properly. It gave correct results... But still couldn't figure out what is wrong when I give those samples of trees

chameerawijebandara commented 9 years ago

OK machan,

What happen if we feed letter 'A' to your code?

On 10 August 2015 at 01:04, HBMDDS notifications@github.com wrote:

I implemented the Neural Network code to recognize hand writings of '1' and '0' using some sample images of hand writing in internet. I did it to confirm whether my code is working properly. It gave correct results... But still couldn't figure out what is wrong when I give those samples of trees

— Reply to this email directly or view it on GitHub https://github.com/chameerawijebandara/Satellite-Image-Analyzer-/issues/3#issuecomment-129228187 .

HBMDDS commented 9 years ago

I don't think it will give correct result.. I'll try it

HBMDDS commented 9 years ago

NN is working properly with individual test images. But it's output for an image is not good. It detect lot of unnecessary points as tress...

chameerawijebandara commented 9 years ago

I didnt got it

On Saturday, August 15, 2015, HBMDDS notifications@github.com wrote:

NN is working properly with individual test images. But it's output for an image is not good. It detect lot of unnecessary points as tress...

— Reply to this email directly or view it on GitHub https://github.com/chameerawijebandara/Satellite-Image-Analyzer-/issues/3#issuecomment-131318792 .

HBMDDS commented 9 years ago

If we give 7X7 images of trees and some other images NN will give a good prediction. But, if we use the NN for a large image (eg: 600X400) to detect trees it will detect lot of unnecessary points result

chameerawijebandara commented 9 years ago

Plz add more false tranning samples and try

On Saturday, August 15, 2015, HBMDDS notifications@github.com wrote:

If we give 7X7 images of trees and some other images NN will give a good prediction. But, if we use the NN for a large image (eg: 600X400) to detect trees it will detect lot of unnecessary [image: result] https://cloud.githubusercontent.com/assets/11896142/9289273/068ae68a-4388-11e5-95c6-ed9f6ce8894a.JPG points

— Reply to this email directly or view it on GitHub https://github.com/chameerawijebandara/Satellite-Image-Analyzer-/issues/3#issuecomment-131387357 .

HBMDDS commented 9 years ago

When we maually crop trees and non tree images NN can classify them correctly as trees and non trees, but the problem is how to crop tree images. That has been very difficult. We have be very precise when we crop tree images. Since the trees are very small f we do not crop the properly NN will no detect them correctly. This is the current result from MatLab code which I have developed to identify locations which have the same shape as trees. It detect about 900 locations. After that we have to crop these circled locations and classify them using the NN. In this image most of the tree locations have detected result1

chameerawijebandara commented 9 years ago

Could you please upload the original image as well

HBMDDS commented 9 years ago

image1

chameerawijebandara commented 9 years ago

Good lets try that

HBMDDS commented 9 years ago

Bug is found.. 94587 points were reduced to 396 points by removing non Tree points. Green boxes show the objects detected as trees, this is the latest result of the code result3

HBMDDS commented 9 years ago

result4

HBMDDS commented 9 years ago

Latest result, White dots are the points detected as trees result8

chameerawijebandara commented 9 years ago

Good Shall we try try to color other trees as well (with separate color)

On 29 August 2015 at 04:25, HBMDDS notifications@github.com wrote:

Latest result [image: result8] https://cloud.githubusercontent.com/assets/11896142/9558673/f616d178-4e05-11e5-9d47-c141628546f5.png

— Reply to this email directly or view it on GitHub https://github.com/chameerawijebandara/Satellite-Image-Analyzer-/issues/3#issuecomment-135906321 .