Wireless underground sensor networks (WUSNs) are gradually being applied to smart agriculture for soil information collection and monitoring of crop growth environments. WUSN can avoid the inconvenience caused by tillage and other machine operation activities on farmland and obtain multi-level and multi-dimensional parameters in the underground soil environment, which is crucial for soil moisture monitoring of crops. However, WUSN has no universally applicable transmission protocol standards in the field. Therefore, the research of different soil compositions on the placement of wireless sensor network nodes can provide scientific guidance to obtain soil moisture information of agricultural fields, which is important for the development of precision agriculture. In this paper, low-power WUSN nodes were designed, based on the modified Frisian transmission model and the complex refractive index Fresnel model. We proposed an adaptive optimization model and also proposed an improved genetic algorithm that automatically adjusts the fusion parameter according to soil and distance factors, making the prediction of signal attenuation under different soil components more accurate. We used the adaptive optimized model for signal prediction, comparing it with the modified Friis prediction model and the complex refractive index Fresnel prediction model. The results showed that the proposed adaptive optimization model with an automatic parameter is convenient to predict the signal attenuation, and the adaptive optimization model made the prediction error stay really low. To compare with other sensors in the soil environment, the temperature of the distributed fiber-optic temperature sensor was tested, which was predicted by the adaptive model. The result shows that the adaptive model is more favorable to the prediction of signal attenuation in WUSN than distributed fiber-optic temperature sensors.