MobileNetV2 ve MobileNetV3 Tabanlı Derin Öğrenme Yaklaşımları ile Cilt Kanserlerinin Sınıflandırılması

EasyChair Preprint no. 8429

Abstract

The ozone layer has been depleted as a result of the damage to the environmental environments in which people live due to different reasons. As a result of this puncture, the amount of exposure to beneficial and harmful rays of the sun increased. As a result of the increase in the amount of time people stay under the sun's rays, there are changes in skin colors in various parts of the body. The medical world, on the other hand, defines the changes that occur on the skin in general as skin cancer. Among the people, the most common skin cancer types are known melanoma (mel), dermafibroma (df), vascular (vasc), benign keratosis (bkl), melanocytic nevi (nv), basal cell carcinoma (bcc), actinic keratosis (akiec). Not all changes on the skin are cancerous. In this case, although it is not possible to characterize all changes in the skin as skin cancer, it is possible to determine which image belongs to which cancer using images obtained from skin changes. In this study, the classification success of deep learning models based on transfer learning was compared, considering the inadequacy and workload of skin health specialists. MobileNet V3 Large with a new MobileNet architecture activation function and MobileNet V2 with the ReLU activation function are effectively compared. According to the KFold 2 option, the training and test data were divided into two, and both models, which were used with weighted values in experimental studies, reached training and test accuracy values close to each other.

Keyphrases: deep learning, MobileNet V3, MobileNet-V2, Skin Cancer, Transfer Learning

@Booklet{EasyChair:8429,
  author = {Halit Çetiner},
  title = {MobileNetV2 ve MobileNetV3 Tabanlı Derin Öğrenme Yaklaşımları ile Cilt Kanserlerinin Sınıflandırılması},
  howpublished = {EasyChair Preprint no. 8429},

  year = {EasyChair, 2022}}