Protection against attacks on human gender recognition systems
Jazyk: | ruština |
---|---|
Rok vydání: | 2023 |
Předmět: | |
DOI: | 10.18720/spbpu/3/2023/vr/vr23-812 |
Popis: | Тема вÑпÑÑкной квалиÑикаÑионной ÑабоÑÑ: «ÐаÑиÑа Ð¾Ñ Ð°Ñак на ÑиÑÑÐµÐ¼Ñ ÑаÑÐ¿Ð¾Ð·Ð½Ð°Ð²Ð°Ð½Ð¸Ñ Ð¿Ð¾Ð»Ð° Ñеловека».ЦелÑÑ ÑабоÑÑ ÑвлÑеÑÑÑ ÑеализаÑÐ¸Ñ Ð¼ÐµÑода заÑиÑÑ Ð¾Ñ Ð°Ñак на ÑиÑÑÐµÐ¼Ñ ÑаÑÐ¿Ð¾Ð·Ð½Ð°Ð²Ð°Ð½Ð¸Ñ Ð¿Ð¾Ð»Ð° Ñеловека. ÐÑедмеÑом иÑÑÐ»ÐµÐ´Ð¾Ð²Ð°Ð½Ð¸Ñ ÑвлÑÑÑÑÑ Ð°Ñаки на нейÑоÑеÑевÑе меÑÐ¾Ð´Ñ Ð¾Ð¿ÑÐµÐ´ÐµÐ»ÐµÐ½Ð¸Ñ Ð¿Ð¾Ð»Ð° Ñеловека, оÑновÑваÑÑиеÑÑ Ð½Ð° деÑекÑии и клаÑÑиÑикаÑии лÑдей. ÐадаÑи, ÑеÑаемÑе в Ñ Ð¾Ð´Ðµ иÑÑледованиÑ:ÐÑÑледование пÑинÑипов ÑÑÑеÑÑвÑÑÑÐ¸Ñ Ð°Ð»Ð³Ð¾ÑиÑмов и нейÑоннÑÑ ÑеÑей Ð´Ð»Ñ Ð·Ð°Ð´Ð°Ñ ÐºÐ»Ð°ÑÑиÑикаÑии по полÑ.ÐÑÑледование ÑÑенаÑиев аÑак на ÑиÑÑÐµÐ¼Ñ ÑаÑÐ¿Ð¾Ð·Ð½Ð°Ð²Ð°Ð½Ð¸Ñ Ð¿Ð¾Ð»Ð° Ñеловека.Создание даÑаÑеÑа Ð´Ð»Ñ Ð¾Ð±ÑÑÐµÐ½Ð¸Ñ Ð¸ ÑеÑÑиÑÐ¾Ð²Ð°Ð½Ð¸Ñ Ð½ÐµÐ¹Ñонной ÑеÑи.РеализаÑÐ¸Ñ Ð¼ÐµÑода заÑиÑÑ Ð¾Ñ Ð°Ñак на ÑиÑÑÐµÐ¼Ñ ÑаÑÐ¿Ð¾Ð·Ð½Ð°Ð²Ð°Ð½Ð¸Ñ Ð¿Ð¾Ð»Ð° Ñеловека.Ð Ñ Ð¾Ð´Ðµ ÑабоÑÑ Ð¸ÑÑÐ»ÐµÐ´Ð¾Ð²Ð°Ð½Ñ ÑÑÑеÑÑвÑÑÑие алгоÑиÑÐ¼Ñ Ð¸ нейÑоннÑе ÑеÑи Ð´Ð»Ñ Ð·Ð°Ð´Ð°Ñ ÐºÐ»Ð°ÑÑиÑикаÑии по полÑ. ÐÑли пÑоанализиÑÐ¾Ð²Ð°Ð½Ñ ÑÑенаÑии аÑак на ÑиÑÑÐµÐ¼Ñ Ð´ÐµÑекÑии обÑекÑов.Ð ÑезÑлÑÑаÑе ÑабоÑÑ Ð¿Ð¾ÑÑÑоена Ð¼Ð¾Ð´ÐµÐ»Ñ ÑвеÑÑоÑной нейÑонной ÑеÑи Ð´Ð»Ñ ÑаÑÐ¿Ð¾Ð·Ð½Ð°Ð²Ð°Ð½Ð¸Ñ Ð¿Ð¾Ð»Ð° Ñеловека, заÑиÑÑÐ½Ð½Ð°Ñ Ð¾Ñ Ð°Ñак вÑедоноÑнÑÑ Ð¿Ð°ÑÑей, бÑла пÑодемонÑÑÑиÑована ÑÑÑекÑивноÑÑÑ Ð´Ð°Ð½Ð½Ð¾Ð¹ модели.ÐолÑÑеннÑе ÑезÑлÑÑаÑÑ Ð¼Ð¾Ð³ÑÑ Ð±ÑÑÑ Ð¸ÑполÑÐ·Ð¾Ð²Ð°Ð½Ñ Ð² каÑеÑÑве оÑÐ½Ð¾Ð²Ñ Ð´Ð»Ñ Ð¿ÑоекÑиÑÐ¾Ð²Ð°Ð½Ð¸Ñ Ð·Ð°ÑиÑеннÑÑ ÑиÑÑем ÑаÑÐ¿Ð¾Ð·Ð½Ð°Ð²Ð°Ð½Ð¸Ñ Ð¿Ð¾Ð»Ð° Ñеловека. The topic of the graduate qualification work is «Protection against attacks on human gender recognition systems».The purpose of the study is to implement a method of protection against attacks on human gender recognition systems. The subject of the work is attacks on neural network methods of human gender identification relying on face detection and classification. The research set the following goals:Studying the principles of existing algorithms and neural networks for gender classification tasks.Research of attack scenarios on human gender recognition systems.Creating a dataset for training and testing a neural network.Designing a method for protecting against attacks on human gender-recognition systems.During the work, existing algorithms and neural networks for gender classification tasks were studied. Scenarios of attacks on object detection systems were analyzed.The work resulted in the construction of a convolutional neural network model for human gender recognition, protected from attacks of adversary patches, and demonstrated the effectiveness of the model.The results can be used as a foundation for the design of protected human gender recognition systems. |
Databáze: | OpenAIRE |
Externí odkaz: |