Electronic Theses and Dissertation

Wajah memiliki karakteristik tertentu yang menyimpan informasi penting mengenai usia sesorang. Usia dapat diklasifikasikan dengan mengamati gambar secara visual dan mengidentifikasi wajah berdasarkan fitur seperti jarak mata, panjang hidung, jarak bibir, dan lain- lain. Manusia memiliki kemampuan mengelompokkan usia seseorang hanya dengan melirik atau melihat wajah seseorang. Namun, manusia juga memiliki keterbatasan indra penglihatan, sehingga tidak menutup kemungkinan manusia bisa saja keliru dalam mengklasifikasi usia seseorang. Penelitian ini bertujuan untuk membangun model deep learning dengan dataset yang sama, namun jenis citra yang berbeda yaitu citra grayscale dan citra image sharpening dalam mengklasifikasi usia. Model ini menggunakan dua arsitektur yaitu EfficientNetB0 dan ResNet-50. Hasil pengujian menunjukkan bahwa model yang dibangun memiliki performa yang baik. Pada citra grayscale
dengan arsitektur EfficientNet-B0 dengan epoch 200 menghasilkan accuracy sebesar 88,04%, precision 88,77%, recall 88,04%, dan f1-score 87,62%. EfficientNet-B0 dengan epoch 300 accuracy sebesar 85,71%, precision 86,82%, recall 85,71%, dan f1-score 84,41%. EfficientNet-B0 early stop menghasilkan accuracy sebesar 80,18%, precision 80,25%, recall 80,18%, dan f1-score 79,61%. ResNet-50 menghasilkan accuracy sebesar 88,39%, precision 89,49%, recall 88,39%, dan f1-score 87,39%. Pada citra sharpening dengan arsitektur EfficientNet-B0 epoch 200 menghasilkan accuracy sebesar 88,21%, precision
88,76%, recall 88,21%, dan f1-score 88,61%. EfficientNet-B0 dengan epoch 300 accuracy sebesar 86,79%, precision 88,33%, recall 86,79%, dan f1-score 85,41%. EfficientNet-B0 early stop menghasilkan accuracy sebesar 80,18%, precision 81,19%, recall 80,18%, dan f1-score 79,14%. ResNet-50 menghasilkan accuracy sebesar 88,57%, precision 89,22%, recall 88,57%, dan f1-score 88,05%. Dengan menggunakan algoritma pre-processing image sharpening, citra yang dihasilkanmemiliki ketajaman yang tinggi dan terlihat lebih baik dalam menampilkan fiturpada citra sehingga mampu menghasilkan model performa yang lebih baik

Faces have certain characteristics that hold important information about a person's age. Age can be classified by visually observing images and identifying faces based on features such as eye distance, nose length, lip distance, etc. Humans have the ability to determine someone's age just by glancing or looking at someone's face. However, humans also have limited sense of sight, so it is possible that humans could be wrong in classifying someone's age. This research aims to build a deep learning model with the same dataset, but different types of images, namely grayscale images and image sharpening images in classifying age. This model uses two architectures, namely EfficientNetB0 and ResNet-50. The test results show that the model built has good performance. Grayscale images with the EfficientNet-B0 architecture with epoch 200 produce accuracy of 88.04%, precision of 88.77%, recall of 88.04%, and f1-score of 87.62%. EfficientNet-B0 with epoch 300 accuracy of 85.71%, precision 86.82%, recall 85.71%, and f1-score 84.41%. EfficientNet-B0 early stop produces an accuracy of 80.18%, precision of 80.25%, recall of 80.18%, and f1-score of 79.61%. ResNet-50 produces accuracy of 88.39%, precision of 89.49%, recall of 88.39%, and f1-score of 87.39%. Image sharpening with theEfficientNet-B0 epoch 200 architecture produces an accuracy of 88.21%, precision of 88.76%, recall of 88.21%, and f1-score of 88.61%. EfficientNet-B0 with epoch 300 accuracy of 86.79%, precision 88.33%, recall 86.79%, and f1-score 85.41%. EfficientNet-B0 early stop produces an accuracy of 80.18%, precision of 81.19%, recall of 80.18%, and f1-score of 79.14%. ResNet-50 produces an accuracy of 88.57%, precision of 89.22%, recall of 88.57%, and f1-score of 88.05%. By using a pre-processing image sharpening algorithm, the resulting image has high sharpness and looks better in displaying features in the image so as to produce a better performance model.