Ensemble of weighted deep concatenated features for the skin disease classification model using modified long short term memory

Abstract

Skin diseases are considered to be a common disease in human, which have many invisible dangers that may reduce the self-confidence and causes certain psychological depression and in-depth, these skin diseases may also results in skin cancer. These skin diseases needs to be diagnosed with the medical experts but it requires high-level instruments for diagnosing as they suffer from the visual resolution problem while analyzing the skin disease images. Hence, it is important to implement a computer-aided detection scheme for automatically diagnosing the skin disorder. Hence, this work plans to implement an effective skin disease classification method with the help of novel deep learning methodology. Initially, the dataset is gathered and pre-processed by the contrast enhancement technique through “histogram equalization”. After pre-processing, the segmentation of the images is done by the Fuzzy C Means segmentation (FCM). Further the segmented images assigned as a input for the deep feature extraction using Resnet50, VGG16, and Deeplabv3. The features are attained from the final layer of these three techniques and are concatenated. These concatenated features are provided to the feature transformation phase through the weighted feature extraction, which is performed by Hybrid Squirrel Butterfly Search Optimization (HSBSO). The transformed features are given to Modified Long Short Term Memory (MLSTM), where the architecture optimization is done by the same HSBSO for producing the final classified output. The analysis results confirms that the better effectiveness of the introduced method when observing the accuracy of the implemented skin disease classification than the conventional approaches.