Plastic object detection with an infrared hyperspectral image


Thesis Type: Postgraduate

Institution Of The Thesis: Middle East Technical University, Graduate School of Natural and Applied Sciences, Turkey

Approval Date: 2019

Thesis Language: English

Student: MEHMET FATİH DİRİ

Principal Supervisor (For Co-Supervisor Theses): Mehmet Lütfi Süzen

Co-Supervisor: Koray Kamil Yılmaz

Abstract:

Undoubtedly, a world without plastic, which is the most important production material in almost every area, seems inconceivable today. The production and consumption chain, by which induced this versatile use, has caused plastic pollution that has devastating effects on the environment and natural ecosystems. In this thesis, which written with the motivation of contributing to the fight against plastic pollution and to be useful developing effective and sustainable policies, plastic pollution and pollutant types have been investigated; plastic objects have been examined in terms of physical, chemical and spectral aspects; and have been detected on land with an unsupervised manner through shortwave infrared hyperspectral image. 15.5-meter resolution 224 band hyperspectral image which was acquired by AVIRIS is used. In this study, 15 different study field, each of which includes significant plastic object samples like a greenhouse, an artificial turf football pitch, a solar panel, and a tent, is determined within the image scene. The positive value of spectral absorption around 1.72 µm, which is associated with the presence of plastic, has been mathematically expressed using two neighboring shoulders. This algorithm, which has the capability of detecting plastic objects on land quickly and precisely without needing any reference data and using only 3 shortwave infrared bands, has been named as Plastic Existence Index (PEI). The positive values generated as a result of the algorithm has been called Post Index Positive Value (PIPV). Since it was not possible to collect any data from the ground, the reference data has been produced by visual inspection method on the true color composite AVIRIS image. After implementation results have been compared with reference data, it is seen that highly-satisfactory outcomes have been obtained which mean value of UA is 90.51%, PA is 89.04% and OA is 97.37.