Hits:
Affiliation of Author(s):电气与电子工程学院
Journal:Genetic and Evolutionary Computing - Proceedings of the Fourteenth International Conference on Genetic
Funded by:企事业单位委托科技项目
Key Words:target recognition，Faster RCNN network，RPN ROI align
Abstract:Aiming at the problem of missed detection in target detection under special conditions such as occlusion, an improved Faster-RCNN target detection algorithm is proposed. First replace the VGG16 network with Resnet101 and optimize it. Secondly, improve the RPN module, extract features from feature layers with richer semantics, and use two 3 × 3 convolutions to obtain 5 × 5 convolution effects, while reducing training parameters and improving the detection accuracy of occluded targets; finally Improve the RoI pooling to RoI Align, cancel the quantization operation, and use the bilinear interpolation method to obtain the image values with the coordinates as floating-point numbers, thereby transforming the feature aggregation process into a continuous operation. The actual measurement results show that the algorithm is suitable for detecting occluded objects in different scenes and has good robustness.
Co-author:尹博睿,初明
First Author:chengfangxiao
Indexed by:Essay collection
Page Number:2
Translation or Not:no
Date of Publication:2022-04-12