DeepLearningBrickRecognition

A deep learning project based on TensorFlow that recognizes color patterns of brick.

Problem definition

A model needs to learn recognizing following color patterns:

In order to do this I converted all image dataset (about 4000 pictures) into histograms with bins from range 0-255. Here is a quick explanation what is "bin":

• "A histogram displays numerical data by grouping data into "bins" of equal width. Each bin is plotted as a bar whose height corresponds to how many data points are in that bin."

So I decided to take these data points from vertical bars and put them in .csv file.

• Every column represents one bin
• Every cell in row represents one value from that bin
• Every row represents new image

So let's say we take first cell in this dataset:

and this value tells us that in the first picture on bin with number 0 we have 31604 data points corresponding to that bin.

At this point we can feed our network and shuffle the data using "sample" method from pandas library.

Results

Train accuracy: 94%

Test accuracy: 93%