Landslides are a common geological disaster, and due to their sudden and destructive power, often pose a serious threat to human life and property safety, so the establishment of an accurate landslide disaster real-time monitoring and early warning system is very important. In this paper, a landslide monitoring and early warning system are designed with STM32F103 as the core controller, and the main influencing factors of landslides are collected: rainfall, pressure, displacement, and soil moisture content, and the data is transmitted to the on-site early warning terminal by GPRS wireless communication to determine whether the set threshold is exceeded if it exceeds the immediate alarm, it will be transmitted to the remote control center for analysis and processing, and the control center will input the data into the dung beetle algorithm (Dung Beetle Optimizer, DBO) The optimized BP neural network predicts the probability of the current landslide, and divides the landslide warning level according to the probability prediction results, so as to realize real-time landslide monitoring and early warning. Through the comparison experiment between SVM, BP, GA-BP, SSA-BP models, and DBO-BP models, it is concluded that DBO-BP prediction accuracy is higher, and its goodness-of-fit is 98.8%, which is closer to the real value, and compared with the expensive cost of Beidou, GPS, and other technologies in landslide warning, the cost reduced by the embedded landslide disaster monitoring and early warning system has certain engineering application value.