现代电子技术2025，Vol.48Issue(13)：57-61,5.DOI:10.16652/j.issn.1004-373x.2025.13.008
基于级联式逆残差网络的游戏图像多模态目标精准辨识研究
Research on game image multimodal object precise identification based on cascaded inverse residual network
摘要
Abstract
The objects in the game have complex features such as shape,color and texture.In addition,the objects appear in different perspectives,scales and postures,all of which increase the difficulty of object recognition.A cascaded inverse residual network can enhance the multidimensional features of the object,and can recognize the object effectively even in the presence of occlusion,deformation,etc.Therefore,a game image multimodal object precise identification method based on cascaded inverse residual network is proposed.A backbone network consisting of convolutional layers and cascaded inverse residual modules based on depthwise separable convolution design is constructed.This network is used to extract the inputted game image features preliminarily.The channel rearrangement is used to enhance the information exchange among channels.The feature enhancement network is used to upsample the feature maps learned by the backbone network.The multimodal object features are extracted in combination with the multi-channel feature fusion.The object position,direction,and other information are outputted by a prediction network that can achieve classification and regression tasks.So far,a precise identification of multimodal objects in game images is achieved.The experimental results show that the method can achieve the identification of the characters,text,scene elements and other objects in the game images,with a training loss of only about 0.05 and an F1-score of 0.967.To sum up,the multimodal object recognition effect of game images is good.关键词
逆残差/游戏图像/多模态/通道重排/SIoU损失函数/目标辨识/卷积层/目标位置Key words
inverse residual/game image/multimodal/channel rearrangement/SIoU loss function/object identification/convolutional layer/object location分类
电子信息工程引用本文复制引用
刘建志..基于级联式逆残差网络的游戏图像多模态目标精准辨识研究[J].现代电子技术,2025,48(13):57-61,5.基金项目
2023年度辽宁省教育厅基本科研项目(JYTQN2023048) （JYTQN2023048）
