92
S. Zorgui et al.
3 Proposed Method for Lentigo Detection
The proposed method consists of classifying RCM images into healthy/lentigo
classes using an InceptionV3 architecture. Our lentigo detection method combines the InceptionV3 model with other known deep learning techniques like
transfer learning [18] and data augmentation [19]. Figure 3 presents the different
steps used in the proposed lentigo detection method. The following subsections
give detailed descriptions of each step.
Fig. 3. Pipeline of the proposed method.
3.1 Data Preparation
The input RCM images for the training procedure combines two sets such as
a 73% training set and a 14% validation set. The remaining 13% is dedicated
to the prediction phase. In order to avoid overfitting, a validation set is added
to our training phase because the non linear InceptionV3 model will possibly
achieve 100% training accuracy and overfit.
S. Zorgui et al.
3 Proposed Method for Lentigo Detection
The proposed method consists of classifying RCM images into healthy/lentigo
classes using an InceptionV3 architecture. Our lentigo detection method combines the InceptionV3 model with other known deep learning techniques like
transfer learning [18] and data augmentation [19]. Figure 3 presents the different
steps used in the proposed lentigo detection method. The following subsections
give detailed descriptions of each step.
Fig. 3. Pipeline of the proposed method.
3.1 Data Preparation
The input RCM images for the training procedure combines two sets such as
a 73% training set and a 14% validation set. The remaining 13% is dedicated
to the prediction phase. In order to avoid overfitting, a validation set is added
to our training phase because the non linear InceptionV3 model will possibly
achieve 100% training accuracy and overfit.