Citizen science has emerged as an effective approach for infectious disease surveillance. With advancements in machine learning, entomologists can now be relieved from the labor-intensive task of species identification. However, deploying machine learning models on mobile devices presents challenges due to constraints on battery life and memory capacity. In this study, we explore the potential of various model compression techniques for deploying machine learning models on resource-limited devices, enabling low-energy consumption and on-device processing for disease surveillance in remote or low-resource settings. We compared two main-stream model compression techniques, pruning and quantization on various mobile devices. Our findings indicate that quantization methods outperform pruning methods in terms of efficiency. Furthermore, we propose to integrate structured and unstructured pruning to enhance model performance while addressing key constraints of mobile deployment.