Abstract:
Aimed to mitigate the challenges associated with manual detection, such as elevated risks and limited accuracy, a machine vision-based automatic measurement method is pro...Show MoreMetadata
Abstract:
Aimed to mitigate the challenges associated with manual detection, such as elevated risks and limited accuracy, a machine vision-based automatic measurement method is proposed for determining the area of a ring cooler corner. Specifically, an enhanced lightweight model known as EU-Net (Efficient U-Net), derived from the classic semantic segmentation U-Net model, is developed to accurately segment the ring cooler corners within the images. Subsequently, the OpenCV toolkit is employed to conduct area statistics on the segmented masks. The results show that EU-Net achieves comparable segmentation effectiveness and area statistics results to the classic U-Net, while utilizing only 0.097% of the parameters present in the latter model. Our improved algorithm obtains 99.74% and 94.5% in pixel-wise accuracy and intersection over union. Moreover, it can process 60 frames per second, effectively meeting real-time operational demands.
Published in: 2023 2nd International Conference on Artificial Intelligence, Human-Computer Interaction and Robotics (AIHCIR)
Date of Conference: 08-10 December 2023
Date Added to IEEE Xplore: 29 April 2024
ISBN Information: