An Innovative Software Development Methodology for Deep Learning-Driven Visual Computing in Built Environment Applications

This paper presents an innovative software development methodology, the GENESIS (Generalised ENgineering for Embedded Software with Integrated AI System) Methodology, tailored for Deep Learning (DL)-driven visual computing applications in the built environment. Integrating AI into embedded systems has presented unique challenges to the associated software development methodologies. The proposed GENESIS Methodology integrates Design Science Research principles with established Artificial Intelligence (AI) embedded software-specific software engineering practices. Further, the approach has co-opted and synthesised insights from recent studies on AI software development and software engineering methodologies, incorporating key elements. The GENESIS Methodology encompasses twelve key stages, from problem definition to monitoring and maintenance for the developed software systems, with the sharing of knowledge, focusing on data-centric development and model-driven AI approaches. The systematic integration of AI-specific software engineering stages within conventional software engineering methodology uniquely combines a research-driven approach. The emphasis on the importance of Convolutional Neural Networks (CNNs) for visual computing tasks details the technical considerations for training and evaluating Deep Learning models. The paper justifies adopting the Waterfall model for its structured approach, aligning with the needs of the technically complex systems. Finally, a software prototype development is presented using the proposed GENESIS Methodology, and the functionality is focused on the built environment, validated by achieving a 91.2% accuracy in identifying six types of concrete defects, demonstrating the accuracy of this approach in real-world applications. This comprehensive methodology aims to enhance the development of AI-based visual computing applications in the built environment, offering a systematic framework.