Design And Implementation Of Visual Recognition System Based On Shallow Features

After continuous innovation and improvement,computer vision technology has been widely used in the industrial field and greatly improved the efficiency and quality of industrial production.In many industrial scenarios,it is only necessary to identify specific and accurate targets,such as equipment inspection,equipment maintenance assistance and industrial production line item sorting.Considering production efficiency,input cost,recognition accuracy and other factors,the shallow features extraction is more suitable for the recognition of such targets.The shallow features are features extracted by fixed algorithms based on domain knowledge.In the filed of printing,the sorting of printed matter usually requires the identification of specific printed matter.his paper studies the visual recognition method based on shallow features,designs and implements the corresponding recognition system.Aiming at the indentification problem of print sorting in the printing field,this paper designs and implements a visual recognition system based on shallow features.For specific and accurate targets,this paper realizes target recognition from the perspective of image similarity detection.Considering the requirements of print recognition on illumination change,rotation,scale and other aspects,this paper has carried out a large number of comparative experiments on various similarity detection methods based on shallow features,and selected SIFT algorithm as the key algorithm of this system.The features extracted by SIFT calculation have the following properties.First,these features are invariant to the scaling and rotation of the image.At the same time,it alse improves the good matching under the influence of different factors,such as noise and brightness.Second,the number of features is large,even small pictures can also get a large number of features through intensive sampling.The system firstly establishes a small sample image database,which is used to store the sample image,the sample image features,and the description of the sample image.Then,SIFT algorithm was used to extract image features and optimize the process of feature extraction,RANSAC algorithm was used to eliminate image feature mismatching,so as to realize matching and recognition with sample images and meet the requirements of time efficiency.In addition,the system realizes the function of online learning,which can increase the type of system identification samples in the process of system operation.After experimental verification,the system can realizethe identification of print content in real time and accurately.Finally,this pager uses VS2017,OpenCV to realize the function of the system.