Content Detection of Web Pages using HTML2Canvas and YOLOV3

Web Pages contains enormous amount of data that is visible to all types of internet users. Web Pages is built using HTML and it is possible to capture the layout of a web page by converting to a canvas. You only look once algorithm is the best algorithm for object detection. In this study, a YOLOV3 model was created to detect pornographic content, educational content, and gaming content using the Canvas of the web pages saved into JPEG format. The images were annotated using Microsoft Vott. The YOLOV3 model was trained using the pre-trained weights of ImageNet. The accuracy of the YOLOV3 for detecting adult content in pornographic web page canvas has an accuracy of 84% while the educational and gaming web page canvas has an accuracy of 53%. For the safety of readers, the images presented in this paper were blurred or we covered portions of the canvas that showed sexual nudity.