Parameter Fine-Tuning with Transfer Learning for Gender Classification using Palm Images
Gender classification from biometric features has gained significant attention in computer vision and pattern recognition research. This study presents an approach for gender classification using palm images through parameter fine-tuning with transfer learning techniques. Traditional gender classification methods primarily rely on facial features, but palm-based classification offers advantages in scenarios where facial images are unavailable or of poor quality. This study evaluates the effectiveness of parameter fine-tuning with transfer learning techniques for automated gender classification using palm image analysis. We implemented and compared multiple pre-trained convolutional neural network (CNN) architectures including, ResNet-50, VGG-16, and DenseNet-121, applying various fine-tuning strategies on a dataset of 2,400 palm images. The methodology involved progressive unfreezing of network layers, adaptive learning rate scheduling, and comprehensive data augmentation techniques. The fine-tuned ResNet-50 model achieved the highest classification accuracy of 94.7% with precision and recall values of 0.943 and 0.951 respectively. Transfer learning with selective layer fine-tuning demonstrated superior performance compared to training from scratch, reducing computational requirements by 65% while maintaining high accuracy. Conclusions: Parameter fine-tuning with transfer learning provides an effective approach for gender classification using palm biometrics, offering practical applications in forensic science, security systems, and demographic analysis.
Download Paper