In-Vivo Skin Capacitive Image Classification Using AlexNet Convolution Neural Network

© 2018 IEEE. Skin water content is very important for its cosmetic properties and its barrier functions, however, to measure it is very difficult. We have recently developed a novel hand-held probe for in-vivo skin hydration imaging based on the capacitance measurement principle. It is more repeatable, reproducible and easier to calibrate than the existing commercial devices. Our latest research is to assess the performance of deep learning in in-vivo skin capacitive image analysis using AlexNet model. As we know the AlexNet model can be used for image classification and recognition with high accuracy. Our object is to design a model to classify more than one specific features, i.e. not just the one with highest probability. We trained the image classifier using the pretrained model to implement the specific feature extraction, prediction and classification specifically for the skin hydration level, skin damage level and gender. There are over 1000 skin images which are measured by two experiments: repeatability of different instruments in in-vivo skin measurement; and skin damage measurements by different instruments. The objective of the research has been divided into three parts: feature extraction implementation using the pretrained model AlexNet; accuracy assessment of the model; further improve the system for multiple features classification. The image classification programme shows a good result which has accuracy 0.98, and the test images were classified correctly comparing with the experimental results of skin hydration, skin damaged level and the gender of the volunteers.