Collaborate with local NGOs to reduce the amount of plastic wastes in water bodies.

An analytical platform that recieves video feed from Drones to identify and analyze plastic waste on water bodies.

Leverages a pretrained Yolov5 model to predict plastics found on the water surface

It was built with Power apps and Streamlit https://seax.powerappsportals.com/

For training: Dataset from Japan: https://zenodo.org/record/4552389

For testing: Gathered data from a nearby river using drones and plastic bottles.

LabelImg was used to annotate the images.

To propose an efficient way to detect and analyse different plastic types

The paper is divided into two parts

• Detection part
• Analysis part

• Allow for the use of video input

Will be added soon

To acheive the same results from Yolov5 by leveraging a UNet CNN architecture. This research uses the same dataset, preprocessing method but with a UNet model and a extra layer. The results are a displayed differently in that using semantic segmentation to show the mapped areas and a collective accuracy rather than individual predictions.

This is an improvement of a similar poster I presented at Data Scientist Bootcamp 2021 using a different model approach, Efficent Unet. Here's the poster of the previous one.

Refernce: https://github.com/ultralytics/yolov5/blob/master/detect.py https://binginagesh.medium.com/small-object-detection-an-image-tiling-based-approach-bce572d890ca#03ca https://openaccess.thecvf.com/content_CVPRW_2020/papers/w22/Baheti_Eff-UNet_A_Novel_Architecture_for_Semantic_Segmentation_in_Unstructured_Environment_CVPRW_2020_paper.pdf