Incidents involving crop damage and human injuries caused by harmful animals are increasing in Japan. Existing research on detecting harmful animals by utilizing a LiDAR and a rule-based radio wave sensing has been conducted. However, challenges remain in terms of the limitations of range, accuracy, cost-effectiveness, and continuous operation. Therefore, this study proposes a system that enables detection, analysis, and user notification of harmful animals by integrating low-cost, low-power sensor nodes capable of detecting a wide range of harmful animals using radio wave sensing with an unmanned aerial vehicle (UAV). Our sensor nodes utilize an adaptively learned autoencoder to detect harmful animals between the transmitter and receiver nodes of the radio wave with greater accuracy and a larger area than previous studies. Additionally, by collaborating with UAV that observes the correct condition of the field and provides the computational resources, the system can adapt to environmental changes by dynamically updating the machine learning model.