CODEWITHSUNDEEP
pythonneural-network
RyeRai
RyeRai

Reputation: 118

Neural network to detect one class of object only

I'm totally new with neural networks (NN) in python, and I do not know if NN can run in raspberry pi 3? for I think the problem is that NN requires good CPU/GPU performance for training , data transferring and calculation.

So is it possible to train a NN with single class training data? inorder to save CPU/GPU?.

For example I want the system to detect only the sea cucumber in an image. A good answer/explanation or link to any example will be very appreciated. THANKYOU PO

Upvotes: 1

Views: 1899

Answers (2)

Yangguang
Yangguang

Reputation: 1785

Can NN run in raspberry pi 3?

Yes, NN can run on Raspberry Pi 3, like Raspberry Pi 3 running Google TensorFlow (Neural Network) and NodeBots. The better way I recommend is that training the NN on PC, then run the NN with the trained weights on Raspberry Pi. GPU will accelerate the training speed greatly than CPU, but if your NN is not large, CPU can also make the work done.

Is it possible to train a NN with single class training data?

It is possible to train a NN to solve the “one class”(sea cucumber in your case) classification problem. But the train data need to add another type class data that contain samples which are not belong to the target class(sea cucumber). Then with the “sea cucumber” and the “non sea cucumber” two classes data, it is a easy binary classification problem for NN.

EDIT:

Easy way to get start

You can reference this blog:Building powerful image classification models using very little data to start a image classifiction work, which implemented a CNN to get the work done.

Upvotes: 1

Yutao ZHU
Yutao ZHU

Reputation: 88

I'm not sure if I understand your question well.

Single class training data maybe not exist at all. If you want to detect only the sea cucumber, it is the two-classes classification problem right? It is the sea cucumber or not. Yes or no are two classes.

Yes right cool people implement the NN on raspberry pi. But to some extent it's just possible but not efficient. A good GPU will speed up much for training.

A PC is enable to train some small NN.

Upvotes: 1

Related Questions