基于Hog特征的深度学习语义描述方法  

Deep Learning Semantic Description Method Based on Hog Features

作　　者：徐启南 袁霄翔 郭梦晴 XU Qinan;YUAN Xiaoxiang;GUO Mengqing(School of Network Engineering,Zhoukou Normal University,Zhoukou,Henan 466001,China)

机构地区：[1]周口师范学院网络工程学院,河南周口466001

出　　处：《移动信息》2025年第4期339-341,共3页Mobile Information

摘　　要：近年来,深度学习在计算机视觉领域有着广泛的应用。其中,图片语义描述作为重要研究内容之一,具有可观的应用前景。文中通过对语义描述相关技术的调研,提出了一种新的语义识别编解码模型。首先,在编码器端提取图片的Hog传统图像特征。然后,利用ResNet进一步提取传统图像特征,并进行编码。最后,解码器接收来自编码器端的编码特征向量,并对图像的语义描述进行推理。该方法在MSCOCO数据集上实现,与单独使用卷积提取特征方法对比,改进后的模型对于语义描述精确度更高。In recent years,deep learning has been widely applied in the field of computer vision.Among them,image semantic description,as one of the important research contents,has considerable application prospects.Through research on semantic description related technologies,a new semantic recognition encoding and decoding model is proposed in the paper.Firstly,at the encoder end,the Hog traditional image features of the image are extracted.Then,traditional image features are further extracted using ResNet and encoded.Finally,the decoder receives the encoded feature vector from the encoder and infers the semantic description of the image.This method is implemented on the MSCOCO dataset,and compared with using convolution to extract features alone,the improved model has higher accuracy in semantic description.

关 键 词：深度学习 卷积神经网络 解码器 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]