An improved multi-object instance segmentation based on deep learning

DOI: 10.48129/kjs.10879


  • Nawaf Alshdaifat Universiti Sains Malaysia
  • Mohd Azam Osman School of Computer Sciences Universiti Sains Malaysia,11800, Pulau Pinang, Malaysia
  • Abdullah Zawawi Talib School of Computer Sciences Universiti Sains Malaysia,11800, Pulau Pinang, Malaysia



Given the recent breakthroughs in recent years, Deep learning (DL) networks have attracted growing interest and attention by researchers and scholars alike due to its importance in detecting and instance segmentation of objects in an image. An entity's example instance segmentation is a critical problem that requires further analysis. However, given the difficulties in adopting object detection and the instance segmentation approach, this study aims to develop an approach to overcome these issues by proposing a new approach based on the recent DL approach in addition to developing an approach for object instance segmentation. The approach presented in this study consisted of three stages in order to improve the recognition approach. First, adopting a DL approach improves the object's detection in the enhanced ResNet (residual neural network) and connects it with the convolution layer for each ResNet block. Second, improving the localization of multiple objects dependent on the Region Proposal Network (RPN) approach, and third, utilizing a complex instance segmentation approach. This study's findings revealed that the suggested approach using a typical benchmark-image dataset, called the COCO dataset, the experiments are carried out and validated using generic evaluation parameters. The proposed approach's performance is verified and measured against the recent image segmentation approach using object instances. The findings also revealed that in terms of average precision over IoU (AP) threshold measurements using different thresholds, the proposed approach obtained improved results compared to other well-known segmentation approaches.