Implementasi CNN Transfer Learning pada Klasifikasi Citra Wajah untuk Identifikasi Umur dan Gender

Face image classification for identifying age and gender has been widely
applied in biometric systems and security. However, the accuracy of
classification still has opportunities for improvement due to variations in
facial expressions and natural changes caused by aging. The aim of this
research is the implementation of Transfer Learning convolutional neural
networks (CNNs) to enhance the accuracy of face image classification. A
total of 18,591 face images from the Adience Benchmark dataset were
used as the research data. Several Transfer Learning CNN architectures
were tested, including VGG16, InceptionV3, ResNet, and DenseNet.
Image data preprocessing was also conducted by resizing to various
dimensions to analyze its impact on accuracy and computational speed.
The research results indicate that InceptionV3 achieved the highest
accuracy of 76% in gender classification, while ResNet excelled in age
classification with an accuracy of 45%. Overall, the performance of
Transfer Learning CNN in this study did not surpass the combined results
of wide CNN and Gabor filter as in previous studies. Nonetheless, it still
shows improvement compared to traditional methods. Through
hyperparameter optimization and data augmentation, it is expected that the
accuracy of Transfer Learning CNN can be further optimized in the future