Research on Optimization of Object Detection Technology Based on Convolutional Neural Network

Aiming at the optimization problem of object detection technology, an instance segmentation model algorithm based on the improved Cascade Mask R-CNN is proposed. Firstly, information fusion is performed on the mask branches of each cascade stage, so that the mask information of the current stage is...

Full description

Saved in:
Bibliographic Details
Published in2020 13th International Symposium on Computational Intelligence and Design (ISCID) pp. 13 - 16
Main Authors Xue, Yang, Wanjun, Huang, Hongyang, Yu
Format Conference Proceeding
LanguageEnglish
Published IEEE 01.12.2020
Subjects
Cascade Mask R-CNN
Feature extraction
Instance segmentation
Object detection
Object segmentation
Proposals
Shape
Task analysis
Training
Online AccessGet full text

Cover

More Information
Summary:Aiming at the optimization problem of object detection technology, an instance segmentation model algorithm based on the improved Cascade Mask R-CNN is proposed. Firstly, information fusion is performed on the mask branches of each cascade stage, so that the mask information of the current stage is jointly determined by the segmentation results of the previous stage and the current stage. Secondly, GA-RPN is used to replace the original RPN to generate high-confidence and low-density proposal regions. At the beginning of the cascade, the IoU threshold is increased for training, thereby improving the performance of the network model. Finally, the model of this paper is implemented, and compared with the official implementation on the public data set, which verifies the efficiency of improved algorithm.
ISSN:2473-3547
DOI:10.1109/ISCID51228.2020.00010