Anti-Adaptive Harmful Birds Repelling Method Based On Reinforcement Learning Approach

To prevent crop damage from harmful birds, various repelling methods have been studied. However, harmful birds are still causing damage in the orchard by adapting to the repelling device according to their biological characteristics. This paper proposes a method called Anti-adaptive Harmful Birds Repelling (AHBR) that uses the model-free learning idea of the Reinforcement Learning (RL) approach to repell harmful birds that can effectively prevent bird adaptation problems. To prevent adaptation, the AHBR method uses a method of learning the bird's reaction to the available threat sounds and playing them in patterns that are difficult to adapt through the RL approach. We also proposed the Long-term and Short-term (LaS) policy to meet the Markov assumptions that make RL difficult to implement in the real world. The LaS policy enable learning of the actual bird's reaction to the sound of a threat. The performance of the AHBR method was evaluated in a closed environment to experiment real harmful bird such as Brown-eared Bulbul, Great Tit, and Eurasian Magpie captured in orchards. Results obtained from the experiment showed that the AHBR method was on average 43.5% better than the threat sound patterns(One, Sequential, Reverse Sequential, Random) used in commercial products.