AI Optimisers and Malware Detection Using Imagary on the Rasberry Pi

Given the diverse and long-lasting nature of Internet of Things (IoT) hardware, which often results in significant disparities in operating systems, processing power, and RAM, this study evaluates the performance of various AI optimisers for image-based malware detection on Raspberry Pi 4 and 5. The paper discusses the new Raspberry Pi AI HAT and explains the rationale for selecting a lightweight model. Due to the inability of Raspberry Pi devices to handle the training process, which frequently led to crashes, model training was performed on a MacBook M2. However, the Raspberry Pi units were capable of processing batches of 32 images to compare prediction outputs and success rates. This work also details the challenges encountered with different hardware and software variants. Significant findings include the achievement of 100’%’ prediction accuracy with certain optimisers, and the paper presents a comparative analysis of these optimisers' performance.