Harmful Wildlife Detection System Utilizing Deep Learning for Radio Wave Sensing on Multiple Frequency Bands

In recent years, the number of accidents of damage to crops and injures caused by harmful wildlife in various places is increasing in Japan, hence research and development of techniques for observing ecology of the wildlife are attracting attention [1]. The existing observation system is mainly utilizing a camera device and an image processing [2]. However, the camera based system should treat a large capacity of data, hence it is not suitable in a place where a broadband communication line cannot be prepared. Therefore, in this research, we propose a new harmful wildlife detection system that can detect an approach of wildlife by utilizing a radio wave sensing. The proposed system obtains time series data of received signal strength of radio waves transmitted between a transmitter / receiver, and estimate the number/type of the wildlife by analyzing the data by utilizing a deep learning technology. Through the experimental evaluation, it has been clarified that the number / type of the wildlife can be identified to accuracy of higher than 90%.