Personalized icon design model based on improved Faster-RCNN

Abstract

In the digital age, as an important element of visual communication, icons have an increasing demand for personalized design. In order to meet the personalized icon design needs of students, education, management, and design fields, a personalized icon design model based on a faster regional suggestion network is proposed. Firstly, the convolutional neural network is improved to extract the multi-attribute features of icons. The transfer learning is used to optimize model parameter sharing. Then, the improved faster region-Convolutional network model is adopted for object detection, enhancing the ability to classify and recognize icons. The designed method had a recognition accuracy of over 80% in different types of icons. Among different types of icon data, the recognition accuracy of office type icons was the worst, with a recognition accuracy of 81.3%. The recognition accuracy of traffic type icons was the highest, with a recognition accuracy of 98.3%. The model had a processing time of less than 350 ms for different types of icons, with the shortest processing time of 233 ms for social media icons. The research results indicate that the proposed model has high practicality in icon personalized design, and can provide convenient tool support for designers, students, teachers, and users in the field of education management, promoting the popularization and application of personalized icon design.