Journal of Advanced Research in Computer Graphics and Multimedia Technology

Skin Cancer Detection through Image Analysis with a Dual- Architecture Deep Learning Approach

Swetta Kukreja — 2026-05-14

This research paper addresses global skin cancer concerns, emphasising the need for early detection through artificial intelligence (AI), specifically leveraging AlexNet and EfficientNet. The study highlights limitations in traditional diagnostic methods and proposes an AI-driven paradigm shift. In this paper details of the development of a deep learning-based approach have been presented, wherein existing differentiating images are augmented, covering data pre-processing, training, and evaluation with rigorous scientific methodology that utilises the parametric classification of a skin cancer-based lesion or visible marking. Findings emphasise the efficacy of both models, envisioning a future where AI and image-based observation play pivotal roles in early skin cancer detection. The AlexNet architecture receives a 98.9% accuracy, while the EfficientNet B1 provides an accuracy of 88.9%. The conclusion underscores the transformative potential of AI and a productive way of combining architecture to multiply the efficacy in skin cancer detection, leading the way for increased accuracy and accessibility in early diagnosis.

DOI: https://doi.org/10.24321/3051.4266.202601

Analysis of FCN8 and YOLOv8 in Enhancing Road Segmentation for Autonomous Driving

Deepa Parasar — 2026-05-14

In the rapidly advancing field of autonomous driving, achieving precise road segmentation is paramount for accurate environmental perception. This research paper introduces an innovative road segmentation approach, which leverages the FCN-8 architecture with VGG16 as its backbone, further enhanced through the implementation of the sigmoid activation function and the Adam optimiser. In addition, we conduct a comprehensive comparative analysis with the YOLOv8 architecture to provide a holistic view of road segmentation solutions in this context. This research is based on a custom dataset specifically curated to address the unique challenges of autonomous driving scenarios, ensuring that the models are tested against real-world conditions. Our evaluation demonstrates the robustness and efficiency of both approaches in the context of autonomous driving, shedding light on their respective strengths and weaknesses. This innovative methodology not only advances road segmentation accuracy but also significantly contributes to the realm of self-driving technology, fostering safer and more reliable autonomous vehicles in real-world applications.

DOI: https://doi.org/10.24321/3051.4266.202602