جامعة الكوت - University of kut - بحوث

السنة	عنوان البحث	نشر البحث
2025	An Effective Method Based on Pre-trained Networks and Bagging Technique for Ethnicity Recognition	International Journal of Intelligent Engineering and Systems
The recognition and classification of ethnicity are significant research areas in computer vision and machine learning. Recognizing ethnicity plays a crucial role in a variety of applications, including targeted marketing and security issues. Given that one of the main challenges in recognizing ethnicity from facial images is the lack of effective and diverse data, in this research, a new method based on deep learning and bagging techniques is presented to overcome this challenge and increase the accuracy of recognizing different types of ethnicities. In the proposed method, two pretrained networks AlexNet and GoogleNet are used in parallel to extract ethnicity-related features. After feature extraction, the optimal features are selected by minimum-Redundancy-Maximum-Relevance (MRMR) algorithm. Finally, ethnicity classification and recognition are performed using the bagging technique based on the random forest algorithm. In this study, three different databases, BUPT-GLOBALFACE, VGGFace2, and BUPT-Balancedface, which contain face images of people belonging to 4 ethnicities, Asian, African, Indian, and Caucasian, were used to evaluate the proposed model. Based on the simulation results, the average accuracy of our proposed method for recognizing different ethnicities in the VGGFace2, BUPT-GLOBALFACE, and BUPT-Balancedface databases is 99.18, 99.02, and 99.24 percent, respectively, which is improved compared to other compared works.