摘要
针对现有图像分类方法存在多模态特征向量提取完整度较低、图像分类正确率较差等问题,提出了一种基于混合网络模型的多模态图像分类方法。采用正则化操作、对数变换算法、顶帽运算算法和伽马变换算法预处理多模态图像,获取了高质量、高对比度、高清晰度的多模态图像,构建了混合网络模型(复杂网络模型与卷积神经网络模型),利用复杂网络模型提取多模态图像特征向量,通过卷积神经网络模型的多层次结构(卷积层、池化层与全连接层)执行多模态图像分类任务,从而实现了多模态图像分类目标。实验结果表明,在不同实验工况下,提出方法应用后多模态图像特征向量提取完整度最大值为99.03%,多模态图像分类正确率最大值为100%。
A multimodal image classification study based on a hybrid network model is proposed to address theproblems of low completeness in extracting multimodal feature vectors and poor accuracy in image classificationin existing image classification methods. Using regularization operations, logarithmic transformation algorithms,top hat operation algorithms, and gamma transformation algorithms to preprocess multimodal images,high-quality, high contrast, and high-definition multimodal images were obtained. A hybrid network model(complex network model and convolutional neural network model)was constructed to extract multimodal imagefeature vectors using complex network models. The multimodal image classification tasks were performedthrough the multi-level structure of convolutional neural network models(convolutional layer, pooling layer, andfully connected layer), thereby achieving multimodal image classification objectives. The experimental datashows that under different experimental conditions, the maximum completeness of feature vector extraction formultimodal images after the proposed method is 99.03%, and the maximum accuracy of multimodal imageclassification is 100%.
作者
黄矽琳
洪岚
HUANG Xilin;HONG Lan(Experimental Training and Information Technology Center,Liming Vocational University,Quanzhou 362000,China;College of Engineering,Huaqiao University,Quanzhou 362000,China)
出处
《佛山科学技术学院学报（自然科学版）》
CAS
2024年第6期38-45,共8页
Journal of Foshan University(Natural Science Edition)
基金
泉州市社科基金资助项目(2019H01)。
关键词
多模态图像
图像描述
图像分类
混合网络模型
multimodal images
image description
image classification
hybrid network model
