Tree trunks with thick bark can be expected to serve as the potential water storage capacity. Moisture content is one of the most important indicators to monitor the physiological condition of tree trunks in the development of agriculture and forestry. Therefore, it is crucial to accurately obtain the thickness and relative dielectric constant of the layered medium in tree trunks. The growth and health status of trees can be evaluated to optimize the water management strategies. To this end, this study aims to invert the dielectric parameters of tree trunk layered structure using radar detection. Non-invasive detection was realized for the internal structure of tree trunk. The electrical characteristics of different layers were obtained for the medium layers in the tree trunk. According to Snell's law, the physiological structure of tree trunks were combined to construct the propagation model of radar signals in the layered structure of tree trunks. Electromagnetic properties of different media layers in tree trunks were considered, such as the thickness and relative dielectric constant differences of bark, sapwood, and heartwood. These key parameters were then evaluated using radar echo signals. In addition, the accurate inversion was also performed on the dielectric parameters of tree trunks using sparse decomposition. K-SVD dictionary training was combined to process and analyze radar signals. A software defined radio (SDR) platform was used to build a tree trunk detection radar system in the experiment. The accuracy of detection was further improved to easily adjust the operating frequency of the radar, and then modify the signal processing under different environments. Layer stripping algorithm was selected to invert the dielectric constant and thickness of the layered structure in the tree trunks. The echo signals detected by radar were collected to evaluate the performance of the improved model. A comparative experiments were conducted on the radar detection under two conditions: aliasing and no aliasing in the echo signal. In the absence of aliasing, the inversion errors of the relative dielectric constant and layer thickness of the tree trunk were within 2.93% and 3.5%, respectively; In the case of echo aliasing, the inversion errors of relative dielectric constant and thickness were 7.52% and 7.61%, respectively. The experimental results show that the robustness and practicality were achieved to accurately estimate the thickness and relative dielectric constant of the layered medium in tree trunks under different signal environments. Application prospects were also offered to promote the plant physiology. The finding can provide the accurate data support to monitor the tree health status. Effective technical references can also be provided to promote the garden management and water control in the sustainable agriculture and forestry.