Page 206 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

(continued)

Item Details
Model Training and The AI model is initially pretrained on the MS COCO dataset to estab-
Fine-Tuning lish general object detection capabilities. It is then fine-tuned using
domain-specific data collected from real-world waste materials, includ-
ing cans, beer cans, plastics, and paper. For regional adaptation, such
as in Canada, fine-tuning further incorporates localized data to enable
fine-grained differentiation between non-alcohol cans, alcohol cans, US
cans, and bi-metal cans.
An operational feedback loop continuously gathers new field data to
support periodic model updates and performance improvements.

Testbeds or Pilot https:// www .bisco .or .kr/ undershop/ 03 _jag/ ag01 .asp [5]
j
Deployments https:// www .sedaily .com/ NewsView/ 29X575U740 [6]

2 Use Case Description

2�1 Description

Plastics are produced from crude oil, and paper is made from pulp. However, it is not sustainable
to continuously consume crude oil and pulp for the use of plastics and paper in our daily lives.
Therefore, we must realize circularity, where plastics are recycled into plastics and paper into
paper.

To achieve true resource circularity, a sufficient amount of high-quality raw materials must be
collected. For this, it is necessary to enable easy recycling (STEP I) and precise sorting (STEP
II). Waste must be separated into a form that is as close as possible to its pure raw material
state to be recyclable. If precise sorting becomes possible, a sustainable circular economy can
be realized.

In one region of Japan, citizens sort waste into as many as 45 different categories, achieving
a recycling rate of over 80%. However, it is unrealistic to expect every citizen to follow such a
complex sorting system. As a result, most regions simplify their regulations, but compliance
rates still remain low. Deposit-refund schemes have been introduced to improve collection
rates for recyclables, but the high processing costs due to manual labor remain a significant
problem.

Our goal is not only to increase the recycling rate but also to secure high-purity recyclable
materials and to realize true resource circularity. Through artificial intelligence (AI), we aim to
achieve this by enabling accurate, rapid, uninterrupted sorting and dramatically reducing costs
through automation. WIMPLE, installed at Busan Facilities Corporation, has made recycling
collection possible in public spaces where it was previously considered impossible. (STEP I)

Waste generated in public places was not only difficult to separate manually afterward but
was also often collected already contaminated, rendering separation meaningless. By using
WIMPLE to separate recyclable waste from general waste directly at the point of disposal,
recycling has been made significantly easier, and the issues of contamination and decline in
recycling value have been substantially improved. Moreover, the data collected from the device
has been utilized to calculate the amount of carbon reduction achieved, thereby raising public
awareness about the value of resource circulation. It has also enabled optimization of collection

170