Score-based mask edge improvement of mask-RCNN for segmentation of fruit and vegetables
Abstract
Machine intelligence based automation plays a significant role in many modern applications, and vision based understanding is a significant element of this. To meet the goals of vision based understanding of images, segmentation plays a vital role through partitioning of regions of interest for further processing. Much research, including basic statistical and modern convolutional neural network based techniques, has been reported for segmentation. However, application based fine tuning is always essential for effective results in complex applications. In this paper, we have proposed a score-based mask edge improvement of Mask-RCNN to segment fruit and vegetable images in a supermarket environment. A modular score-based edge improvement head is proposed for Mask-RCNN to improve the segmentation of fruit and vegetable images. The edge difference between ground truth and estimated edges is filled with a pre-defined score proportional to the pixel-level difference. A cosine similarity based loss function is reduced to improve the edge details following segmentation. A significant improvement has been reported based on the proposed technique.
RAS ID
45369
Document Type
Journal Article
Date of Publication
3-15-2022
Volume
190
Funding Information
Edith Cowan University
Higher Education Comission (HEC)
Pakistan Islamia University of Bahawalpur
School
School of Engineering
Copyright
subscription content
Publisher
Elsevier
Comments
Hameed, K., Chai, D., & Rassau, A. (2022). Score-based mask edge improvement of mask-RCNN for segmentation of fruit and vegetables. Expert Systems with Applications, 190, article 116205.
https://doi.org/10.1016/j.eswa.2021.116205