Low-cost Smart Raven Deterrent System with Tiny Machine Learning for Smart Agriculture

Smart farming is a promising application domain for the Internet of Things, which helps farmers to reduce operating costs and increase their profit with automation. Bird or raven deterrence is one example of the potential use of the Internet of Things paradigm leading to smart farming applications. Recently, drone-based approaches have demonstrated efficacy to detect and expel birds that reduce crop yield, but on the other side, those solutions are relatively high-cost options for farmers because they require expensive devices such as desktop-level computers and autonomous drones. This paper proposes a maintenancefree, energy-efficient, and low-cost smart raven deterrent system with edge computing, which runs completely standalone and self-sustainable. In particular, a tiny convolutional neural network has been proposed and optimized for multi-core microcontrollers. To demonstrate the effectiveness of the system and the neural network, the paper presents a developed prototype system with a novel hexa-core ARM Cortex-M4F platform, namely Spresense from Sony. The evaluation results show that the prototype system obtains a detection accuracy of 77% for test samples and consumes an average power of 85.1 mW.