This paper addressed the issue of livestock crossing lanes. Initially, new positioning and transmission logics were developed for the early warning device designed to prevent livestock from crossing lanes. At the same time, the hardware was optimized by incorporating a solar-powered charging circuit and lithium battery undervoltage protection, thereby extending the operational lifespan of the device. Furthermore, the device was enhanced with auxiliary protection functions. Specifically, its position collection capability was utilized to collect movement data of wild animals, and an improved time-sequence analysis algorithm was utilized to predict their movement trajectories. In this stage, the hardware was further optimized to meet aforesaid requirements. Building on the original approach of timely data transmission following information collection, the design was enhanced to address data integrity and low power consumption, and a TF card was incorporated to temporarily store collected data. The card then uploaded data in a staged manner, thereby reducing the device’s power consumption and extending its operational lifespan. Moreover, an animal migratory trajectory prediction model based on the long short-term memory (LSTM) networks was proposed, incorporating topographic prior information. The gating mechanism was employed to control information flows, featuring three gated modules, including the terrain gate, positioning gate, and memory gate. The iterative data flow was centered around memory states and prediction outputs. The iteration module received inputs of topographic and positioning information in each time step and obtained new memory state and prediction output in accordance with the outcomes captured for the last moment, thus improving the prediction accuracy and reliability. Experimental verification confirms that the project effectively achieves the early warning function for livestock crossing lanes and the location tracking function for2 wild animals. Moreover, compared to traditional LSTM models, VAR time-sequence analysis and Prophet models, the proposed model incorporating topographic prior information exhibits the lowest root mean square error (RMSE) in predicting the longitudinal and latitudinal directions of wildlife migration positions, demonstrating superiority in animal migration trajectory prediction.
Building on existing research, this project was further improved to better realize the early warning function when livestock approach lanes. Meanwhile, the aspect of wild animal protection was further considered to accurately and integrally record animals’ location trajectories. Moreover, a trajectory prediction algorithm incorporating terrains, landforms, and other factors was proposed to effectively predict animal movement patterns.
(1) Tests confirm that the device for the early warning of livestock crossing lanes automatically detects the distance between livestock and roads when the distance exceeds 2m. It then controls the time interval before sending prewarning messages to the driver end in accordance with the detected distance. The closer the animals are to the road, the more frequently prewarning messages are sent to ensure traffic safety.
(2) Upon optimization for low power consumption, the device of trajectory recording can operate continuously in a power supply mode, powered by solar panels and supported by lithium batteries for energy storage. In conditions with extreme weathers, the device can conduct positioning every 5 seconds and send messages every 1 hour based on the power merely supplied by the batteries. It can operate for approximately 30 hours continuously. When configured to perform positioning hourly and transmit messages every 3 hours, it can be used for at least one week.
(3) The trajectory prediction algorithm redesigned in this project achieves higher40 prediction precision compared to traditional LSTM and Prophet models. This enhanced performance offers more valuable and reasonable references for predicting animal migration trajectories and planning protection zones. By incorporating artificial intelligence-based visual technologies, the aforementioned issues will be further investigated in subsequent studies to develop improved solutions for these two issues.
