MatLab deep learning algorithm is employed for color recognition process that provides temperature information of a Polylactic Acid (PLA) based material which shows thermochromic properties. Preprocessing and feature extraction phases of the deep learning algorithm are performed through the Convolutional Neural Networks. Transfer Learning is an effective method in determining classification with a few amounts of information. In this study, the AlexNet Convolutional Neural Network (AlexNet CNN) transfer learning is also applied for color recognition process to provide temperature information.
We have adapted and fine-tuned the AlexNet CNN for our problem. In the algorithm, 20 images for each 30, 32, 34, 36 degrees Celsius respectively and a total of 4 folders (classes) and 80 images were used. To train the network, 64 images were assigned for training and 16 images employed for validation. With the AlexNet CNN network, accuracy of 93,75% validation was achieved. The gradual occurrence of color transitions shows that the material can detect temperature changes between 30°C and 36°C with a specially designed optical sensor. It is also possible to use the obtained parametric color data in making innovative sensor designs via the deep learning and artificial intelligence techniques. The properties exhibited by the material indicate that it can be used in sensitive temperature sensing applications, especially for hypothermia. It is foreseen that the analyzed material could be applied in many areas such as mountaineering, military activities, bio-sensing systems.
Anahtar Kelimeler: Deep learning, color recognition, thermochromism, convolutional neural network, AlexNet
|
