Microplastic Detection in Glass Containers Using Circular Hough Transform and YOLOv8n

This paper serves as a medium to present a novel approach combining Circular Hough Transform in the pre-processing phase with YOLOv8n architecture for microplastic detection in glass containers. Our proposed system is lightweight in nature which processes a dataset of 781 images along with a csv file containing the information of the microplastic fragments present in the respective images, using CPU-based training. The methodology integrates traditional algorithm from the field of computer vision which is the Circular Hough Transform to isolate the necessary region thereby providing a more focused area to work on with a pre-trained neural network from the domain of deep learning to detect and classify microplastic within the samples. The empirical results of the model were as follows: achieved $71.55 \%$ precision, $30.44 \% \mathrm{mAP}$ @ 0.5, and $\mathbf{2 8 . 5 3 \%}$ recall, showcasing a conservative detection behavior suitable for quality control applications. While the performance of the model is not up to the level of GPU-based systems, yet our approach still offers accessibility and deployment efficiency with $\mathbf{8 4 . 7 m s}$ CPU inference time with $\mathbf{2 . 6 9 M}$ parameters.