GPU based Re-trainable Pruned CNN design for Camera Trapping at the Edge

The advent of distributed computing topologies like edge computing has led to an inadvertent paradigm shift in the world of IoT (Internet of Things). Besides overcoming the latency and bandwidth challenges posed by a conventional cloud-based infrastructure, edge computing has brought an obtrusive decline in energy consumption. In this era of proliferating rate of data generation, usage of Deep learning networks has left an astounding impact on the predictive accuracy and processing time of real-time data. This paper proposes a combination of these emergent technologies to solve the challenges faced in camera traps used for wildlife and marine life monitoring. We, further, devise a pruned neural network architecture that uses a Graphical Processing Unit (GPU) to facilitate computation and optimize the hardware. This paper aims at creating a re-trainable deep learning design that brings the processing of captured images closer to the edge and supports relatively higher frame rates.