PYNQ Overlay for SVM-based Melanoma Classification with Learning Ability

Support Vector Machine (SVM), is the most applicable supervised algorithms, used to solve classification and regression problems. Though the SVM classifier is challenging to implement on an embedded system, its benefits of faster processing and lower power consumption make the effort worthwhile. Melanoma is the most lethal form of skin cancer and thus increases mortality rates. Our goal is to create a melanoma-classification system based on the PYNQ Overlay that can be implemented in primary care settings. In contrast to current hardware solutions, which only offer inference capability with regard to melanoma, the proposed system combines learning ability with inference. The PYNQ platform's processing system is used to implement a SVM algorithm for melanoma classification with built-in training capability. The compute-intensive tasks of the proposed algorithm were developed as Intellectual Property (IP) using a high-level synthesis design approach and integrated with the processing system of the heterogeneous PYNQ platform for the hardware component. The experimental results demonstrated greater accuracy in melanoma classification compared to contemporary works, with the added advantage of learning ability.