How do I make my own label data to input into a convolutional neural network to classify oysters?

1 visualización (últimos 30 días)

Jonathan Gundorph Jansen el 22 de Abr. de 2020

0
Enlazar

Enlace directo a esta pregunta

https://la.mathworks.com/matlabcentral/answers/519879-how-do-i-make-my-own-label-data-to-input-into-a-convolutional-neural-network-to-classify-oysters

Editada: Jonathan Gundorph Jansen el 22 de Abr. de 2020

Dear Matlab community,

I am currently trying to implement a Convolutional Neural Network for detecting oysters in images taken from a drone, and get an estimate of the number of oysters in each picture. Please see the attached sample pictures to see what it looks like, where every white patch is an oyster. I want to use Matlabs Deep Learning toolbox to solve this problem, and I have approximately 900 pictures of different locations taken with the drone in one flight path, where the number of oysters and other features vary, and may either lie in clusters or be separated individually.

The problem is, however, that since I collected all the images myself, I have no label data, and as far as I understand I need the label data to train my CNN.

So my question is, how do I create my own label data to accompany each picture? Some people I spoke with suggested using Instance Segmentation or Pixel Segmentation, which I am not familiar with. What do you recommend?

Best Regards

Jonathan