With the continuous development of precision agriculture, the integration of sensor technology, control technology, and agricultural production activities has become increasingly tight. Soil organic matter (SOM) content is an important parameter in precision agriculture, closely related to the growth status of crops. Therefore, in-situ detection of SOM is an important part of variable seeding. Currently, most researchers use spectroscopy to detect SOM. However, the accuracy of spectroscopy is affected by factors such as soil texture, soil moisture content, soil iron oxide content, and soil particle size. Additionally, spectroscopic instruments are expensive and complex. To address these issues, this study designed and developed a real-time detection system for soil organic matter content based on the high-temperature excitation principle. The SOM detection system includes the selection and application of carbon dioxide sensor, the design of detection devices, the design of control systems, and the design of human–machine interface. Benchtop tests showed that the system's response time standard deviation was 0.252 s, indicating good stability. The filtering and cooling effects of the system met the requirements of the detection system. Five different natural soils were collected, and two of them, along with organic soil, were used to create artificially graded soil samples. Multiple linear regression was used for modeling, selecting six models with an R² greater than 0.8 for accurate prediction of the remaining three different soil textures. The modeling results showed that a heating depth of 10 mm was susceptible to external interference, and the accuracy of the model with a heating depth of 15 mm was similar to that of the model with a heating depth of 20 mm. When the heating time exceeded 10 s, both models with different heating depths had an R² greater than 0.8. The accuracy of the models increased with increasing heating time. Among them, the 15 mm-20 s model had the highest accuracy, with an average prediction accuracy of 91.4 %, a maximum accuracy of 94.1 %, and a minimum accuracy of 86.7 %. The experimental results showed that the SOM detection system designed in this study had good predictive capabilities for SOM in different soil textures and could achieve in-situ detection of SOM.