CLASSIFICATION OF ALUMINUM WASTE USING DISCRIMINANT ANALYSIS

New Waste Beverage Cans Identification MethodUNIVERSITAS SRIWIJAYA

23rd September 20141IntroductionWidely used in variety of daily necessities, ex: cans, wrappers medicine, household furnitureHas a major contribution to global climate change.

In 2004, while only 51.2 percent of solid waste were recycled in USA, which is equivalent of energy about 15 million barrels of oil, and in 2012 only 57 percent.

Japan on the other hand has by far the highest level of sustainable waste management, only 2% is landfilled2Recycle of Solid Waste wasteone ton paper from recycled fiber saves approximately 17 treesrecycling one ton of aluminium saves the equivalent in energy of 2,350 gallons of gasoline

3

4Why Solid Waste SortingThe primary challenge in the recycling of solid waste (aluminum or paper) is to obtain raw material with the highest purity. In recycling, highly sorted object stream facilitates high quality end product, and save processing chemicals and energy because various grades of solid waste are subjected to different recycling processes.

6Why Automated Waste SortingThe Solid waste sorting systems are classified into manual and automated systems. In many countries including Indonesia, solid waste are sorted into different grades using a manual sorting system. Manual sorting faces some major problems: laborinconsistentsevere infections 7THE SVS SYSTEM FOR SOLID WASTE SORTING

8

9overview of the system

10

11

12

13EXPERIMENTAL RESULTS TABLE 1: The Ranges of RGB Color Components of the Objects For 100dpi, 200dpi and 300dpi Resolutions RedGreenBlueMin. Max.Min. Max.Min. Max.ABC100161170144155121135NAB100203225199221197221NRC100237240238239250252ABC200172186151171124149NAB200204225201221201217NRC200238241236241249252ABC300179185161168136149NAB300210218207213204212NRC30023624523724624625514TABLE 2: The Minimum and Maximum Ranges of RGB Color Components For Different Grades of The Objects RedGreenBlueMin.Max.Min.Max.Min.Max.161186144171121149203225199221197221236245236246246255Ranges of the color components for Different Grades of The Objects

17Name of Patent/ Industry StandardTechniques Appliedfor IdentificationTypes of SensorFeaturesClassification Success RateTemplate Matching Template MatchingLogitech QuickCam Pro 4000 Web CameraRGB String94.67%Co-occurrence FeaturesRule based ClassifierLogitech QuickCam Pro 4000 Web CameraEnergy for the Co-occurrence matrices90.67%TiTech SystemsNot MentionedNIR, CMYK sensor and color cameraMaterials, shape, color , texture, and four color printing80%MSS SystemsNot MentionedNIR, Color sensor, Gloss, and LigninThe sensor measures the intensity of the materials fluorescence at a specific wavelength in the ultraviolet light.80%Mechatronic Design of a Waste Sorting System for Efficient RecyclingArtificial Neural NetworkFour Sensors: Lignin, Gloss, Stiffness, and Nikon D50 Digital SLR camera as a Color.(i) Average Lignin value, (ii) Gloss meter reading, (iii) Deflection in the upward direction, (iv) Deflection in the downward direction, (v) color variance parameter I, (vi) color variance parameter II.36.6%Fuzzy Inference System Algorithm90.4%The SVS SystemWindow-based subdivision, and Distance Vector with threshold based rules Logitech QuickCam Pro 4000 Web CameraMode and Energy of the RGB Components95.17%The results of the proposed method are compared with results published in literature18CONCLUSIONThe primary emphasis of this work is on the development of a new solid waste identification method for automated sorting systems known as the SVS system bases on Discriminant Analysis. The method described involved window-based subdivision of the image into N-cells, construction of N-candidate template for N-cells, calculation of matching scores of reference templates for the N-cells image, and application of matching score to identify the grade of the object. The SVS system performance for correct object identification is 95.17% with estimated throughput of 21,600 objects per hour with a conveyor belt width of 18. The weight of the throughput depends on the size and grade of the objects.

Another important idea that has been implemented is the adaptability to new subcategories of the primary object grades. The wide range of subcategories of object grades is used to train the system to recognize new subcategories, and as a result the system is scalable and able to provide robust decisions for object grade identification tasks. Besides, the method was trained with many reference templates using different lighting conditions, which overcame the need to maintain lighting consistency during enrollment and identification phases.

21