Home


PET classification method

Amongst home waste they are dominating PET. Many-months observation of household rubbish confirmed it. On that account we decided that one should first select this type of plastic. With additional argument behind it, there is a fact that it is possible again to process PET and to obtain a textile e.g. fleece material. At developing the algorithm developing we decided the simplest and fastest method so it to be possible to use it in the sorting plant of rubbish on the transport belt in the real time.
In the first step, we load the image and converts it to grayscale. The edge detection is then performed to allow the object to be located. A standard mode of operation using the Canny filter was used for the detection of the edge. After locating the object, we calculate the histogram for this part of the image. Analysis of the histogram consists in adding up first hundred, and next hundred last elements of the histogram. Considering the fact that PET is transparent material but the background of images it is black (similarly to the transmission belt in the sorting plant) we are comparing received values of sums. For PET the value of the first sum will be greater, while for other non-transparent materials the second sum will be larger (Fig. 1).

Finally, on this basis, the decision is made to classify the object as PET or non-PET.

Algorithm:
  • load photo
  • convert to gray scale
  • edge detection
  • object localization
  • select object
  • calculate histogram
  • analyse histogram
  • decision: PET / non-PET


    • Fig.1. Comparison of histograms


    Experiment


    The experiment was carried for our database WaDaBa . We used ten sets of data with 200 pictures each that means that we used two thousand images. Table presents results of experiment. These results have preliminary character for developing advanced waste selection techniques based on computer vision techniques.

    Tab.1. Results of experiment
    Name of set No. of images Recogn. rate [\%] FAR [\%] FRR [\%]
    Set A 200 92.0 0 8.0
    Set B 200 92,5 7.0 0.5
    Set C 200 61,0 25.0 14
    Set D 200 74,0 25.0 1.0
    Set E 200 54,0 46.0 0
    Set F 200 88,5 11,5 0
    Set G 200 51,5 48,5 0
    Set H 200 57,0 42.0 1.0
    Set I 200 88,0 11.0 1.0
    Set J 200 80,0 14.0 6.0
    Set K 200 94,0 6.0 0
    Average 75,68 21,45 2,86


    .

    The comparison to other methods using database WaDaBa

    Send us information about your method - according to information in the table below. email: januszb(at)icis.pcz.pl

    No. NameRR [%]FAR [%] FRR [%]Reference
    1Histogram - our 75.6821.452.86 [1]
    2Triple-histogram 81.7 13.5 4.8 [2]
    3Your ? - - [3]

    References

    1. J. Bobulski, J. Piątkowski, PET waste classificationation method and Plastic Waste DataBase WaDaBa, Conference Proc. IP&C 2018, Advances in Intelligent Systems and Computing, vol. 681, Springer Verlag, 2018, pp.57-64. PDF.
    2. Bobulski J., Kubanek M.: The triple histogram method for waste classification, 18TH International Conference of Numerical Analysis And Applied Mathematics ICNAAM 2019, 23-28 September 2019, Rhodes, Greece.
    3. Your

    Home