Pipelined Structure in the Classification of Skin Lesions Based on Alexnet CNN and SVM Model With Bi-Sectional Texture Features

This paper presents a pipelined structure for classifying skin lesions by integrating AlexNet CNN and multi-class SVM models with bi-sectional texture features. The process begins with input image preprocessing using Contrast Limited Adaptive Histogram Equalization (CLAHE), morphological closing, and median filtering to enhance the image quality. AlexNet CNN is then used for feature extraction, with layers trained on a comprehensive skin lesion dataset. The initial classification identifies common skin lesion types: Basal Cell Carcinoma (BCC), Actinic Keratosis (ACK), Melanoma (MEL), Nevus (NEV), and Benign Keratosis (BKL). Following median filtering, K-means clustering and thresholding are applied to segment the lesion, with additional ABCD (Asymmetry, Border, Color, Diameter) features and bi-sectional texture features extracted for enhanced discrimination. These features are classified using a multi-class SVM, further refining the classification of BCC, ACK, MEL, NEV, and BKL. The combination of CNN with AlexNet for feature extraction and SVM for classification, along with the inclusion of bi-sectional texture features, improves accuracy in detecting and classifying various skin lesion types. The proposed pipeline demonstrates effectiveness in automating skin lesion classification and aiding early diagnosis.

Keywords: Skin Disease Dataset, Image Processing Techniques, Deep Learning Techniques, Convolution Neural Network, Classification, machine learning, svm and Accuracy.