DBInV3F-Net: Deep Belief InceptionV3 Fractional Network for Skin Cancer Detection

Skin cancer is a critical health issue, where accurate and early detection is essential to improve survival rates. However, existing diagnostic systems suffer from limited robustness, reduced accuracy, and poor generalization across diverse skin tones, lesion types, and imaging conditions. These limitations significantly affect reliable and scalable clinical decision-making. To tackle these, an innovative approach named Deep Belief InceptionV3 Fractional Network (DBInV3F-Net) is introduced. Initially, the input images are denoised using Regularization Denoising. Segmentation of the skin lesion is carried out by Pelican Optimization Algorithm-based Spatial-Channel Synergistic Optimization Network (POA_SCSONet), where Spatial-Channel Synergistic Optimization Network (SCSONet) is trained using Pelican Optimization Algorithm (POA), and the loss function is modified using the Logarithmic loss. Furthermore, Superpixel-mixing-based image augmentation performs augmentation and significant features are extracted. Lastly, skin cancer is detected by DBInV3F-Net, which is an integration of the InceptionV3 network, Deep Belief Network (DBN), and fractional concept. DBInV3F-Net has attained 96.84% of accuracy, 97.63% of Positive Predictive Value (PPV), and 95.77% of Negative Predictive Value (NPV), and outperforms existing methods such as ConvNeXtV2, EfficientNetV2-M, Extreme Learning Machine-Teaching-Learning-Based Optimization (ELM-TLBO), and Multi-scale GC-T2, with performance improvements of 5.29%, 4.25%, 3.19%, and 2.76%, respectively, considering accuracy.