Geek asks and answers

Image trained artificial neural network

I am interested in creating software that detects an object like a pen using Microsoft Kinect. I recall 100 positive images with 200 negative images so that they can be used by an artificial neural network. My question is: how can I convert these images as ANN input? I assume that the last layer has one neuron, because one output is it or not, and I think that there is only one input, I want to use only 3 layers. But I don’t know if I should convert the positive and negative images to a matrix or what can I do?

+3
source share
1 answer

First of all, welcome to Stackoverflow!

I have never personally considered using Kinect for image recognition, but if possible, you should reduce the image to a fairly reasonable size, such as 100x100to keep it manageable.

You should also try to convert the image to grayscale, as it will also help with computational efficiency, development time, and it is much easier to start with than with RGB.

The input layer will not be equal to 1, which is specified. If we mean an image with a size of 100x100, the total number of inputs should be 10000, one for each pixel. Remember that you are trying to break the data as small as possible so that ANN can detect patterns in the data.

2 neurons . , , . 2 , (, ) (, ). , 2 , , , .

3 , , , . , , ! , .

+4

More articles:


All Articles