基于多模态神经网络的图像弱特征自增强仿真  

Simulation of Image Weak Feature Self enhancement Based on Multimodal Neural Networks

作　　者：夏晶晶[1] 茹广欣[2] XIA Jing-jing;RU Guang-xin(Henan University of Animal Husbandry and Economy Henan Zhengzhou 450046,China;Henan Agricultural University,Henan Zhengzhou 450046,China)

机构地区：[1]河南牧业经济学院,河南郑州450046 [2]河南农业大学,河南郑州450046

出　　处：《计算机仿真》2024年第6期299-303,共5页Computer Simulation

摘　　要：随着计算机技术的快速发展,图像识别技术的应用也日益普遍,对目标图像进行特征增强的需求也进一步增加。为了解决当前图像特征增强算法特征提取能力差、图像噪声大等问题,提出了一种基于注意力机制的多模态神经网络增强算法。算法首先将图像数据、文本描述数据和相似数据集数据作为多模态数据输入,并使用卷积和线性变换使其被调整至同一维度;然后采用神经网络交互模块进行特征融合;随后采用注意力机制模块来加强局部相邻通道间的信息交流,采用池化层模块对目标特征进行增强;最后连接长短期记忆网络得到图像特征输出序列,从而达到特征增强的效果。实验结果表明,所提算法将峰值信噪比提升了8.39%,将边缘保护指数提升了5.15%,提高了弱特征自增强能力。With the rapid development of computer technology,the application of image recognition technology is increasingly widespread,and the demand for feature enhancement of target images is further increasing.In order to solve the problems of current image feature enhancement algorithms such as poor feature extraction ability and high image noise,this paper proposes a multimodal neural network enhancement algorithm based on attention mechanism.The algorithm first inputs image data,text description data,and similar dataset data as multimodal data,and uses con⁃volution and linear transformation to adjust them to the same dimension;Then,a neural network interaction module is used for feature fusion;The attention mechanism module is used to enhance information exchange between local adja⁃cent channels,and the pooling layer module is used to enhance target features;Finally,the image feature output se⁃quence is obtained by connecting the long and short term memory network to achieve the effect of feature enhance⁃ment.The experimental results show that the proposed algorithm improves the peak signal to noise ratio by 8.39%,improves the edge protection index by 5.15%,and improves the self enhancement ability of weak features.

关 键 词：特征增强 多模态 注意力机制 神经网络 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]