基于内容的仪器仪表图像检索系统

Content-based instrument image retrieval system

随着互联网数据的快速增长,基于内容的图像检索系统迅速成为计算机视觉的热点领域,在各行业都起着非常大的实际作用。利用基于图像内容检索相比传统的文本检索更加直观快捷的特点,并且结合设备维保领域的仪器设备检索的实际需求,提出了基于多特征融合和分级检索的图像检索思路,搭建了可用的服务系统,帮助维保工作人员提高工作效率。首先采用了MobileNet-SSD神经网络,训练主体检测模型,对输入图片的主体部分进行裁剪,消除主体以外的其他部分对主体特征提取的影响;然后,通过提取主体图像的最底层颜色特征,形状特征和纹理特征,获得初步筛选结果作为第二级检索的数据集。最后针对发生了旋转或平移的图片,在研究SIFT算法的基础上,利用全仿射变换的算法ASIFT算法进行初步筛选集进行更精确的筛选。实验表明,本算法可以有效地提高图像检索结果的准确率与速度。

With the rapid growth of Internet data, content-based image retrieval system has rapidly become a hot area in the field of computer vision, and plays a very important role in various industries. Based on the characteristics of image retrieval which is more intuitive and faster than traditional text retrieval, and combined with the actual needs of equipment retrieval in the field of equipment maintenance, this paper puts forward the idea of image retrieval based on multr-feature fusion and hierarchical retrieval, and builds an available service system to help maintenance staff improve their work efficiency. Firstly, the MobileNet-SSD neural network is used to train the detection model of the subject, and then the main part of the input image is clipped to eliminate the influence of other parts besides the subject on the extraction of the main feature. Then, the preliminary screening is obtained by extracting the bottom color feature, shape feature. And texture feature of the main image, whose result is used as the data set of the second level retrieval. Finally, aiming at the rotation or translation of the image. Based on the study of SIFT algorithm, the full affine transformation algorithm ASIFT algorithm is used to preliminary screening set for more accurate screening. The experiments show that the algorithm in this paper can improve the accuracy and speed of image retrieval effectively.