Abstract: Abstract: An Energy Dispersive X-Ray Fluorescence (EDXRF) monitor was here developed to detect the soil heavy metal using narrow band internet of things (NB-IoT). The scattered data was also real-time uploaded during detection. The instrument consisted of a TUB00050-AG2 X-ray tube, variable windows collimator, a Vitus H30 40 mm2 detector, an LTC2269 analog to digital converter, and an NB-IoT communication module. The X-ray tube turned on the high voltage and filament current under the control of the main chip, thereby producing bremsstrahlung X-rays, where the electrons bombarded the silver target under a strong electric field. The X-ray was then converted to the object ray with the corresponding peak by the filter collimator. The object ray irradiated the center of the sample by adjusting the divergence angle through the collimator. The fluorescence ray was then reflected on the receiving surface of the silicon drift detector with Compton and Rayleigh scattered rays. The detector converted the photon of the incident ray into the pulse signal for the subsequent step rising signal with the preamplifier. The signal was amplified, held, and sampled to generate the spectrum, and then data and location information were uploaded to the NB-IoT module. The final content of each element was obtained for the spectrum resolution, deviation correction. There were no packets loss, and connections instability during 10 000 times' uploading simulated data, indicating low power consumption and stable signal in the NB-IoT module. Better repeatability, real-time detection, and data integration were achieved in the variable light window collimator and communication means with NB-IoT, compared with other similar devices. The NB-IoT base station can widely be expected to support many devices and cover a large area. A general communication protocol of the Internet of things, MQTT, was set up with an NB-IoT module between the platform and instrument. A wide range of expansion support can realize the integration of multiple instruments and various measurement data. An experiment was also designed to explore the best test time and preheating time of the instrument, where five durations were set. It was found that the instrument performed well, as the duration increased, but some elements became unstable when the duration reached 240 s per sample. The best duration was determined to be 180 s in this case. Consequently, the instrument presented the best repeatability, when the sample was preheated for more than 30 min and the measurement time was 180 s. Three instruments were also fabricated to verify the measurement accuracy of the instrument with the soil samples from the same batch in Sichuan Province. The collected soil was used to prepare the standard samples after drying, grinding, sieving and pressing. This instrument and Olympus Vanta Element-S were compared to measure each sample 5 times. The soil samples were also characterized in a laboratory chemical analysis. It was found that the detection presented a high accuracy for the Cd, Hg, As, Pb, Zn, Cu, and other five elements. Particularly, the measured value of Cr was much more approximate to the true one, compared with Olympus Vanta Element-S. The average relative errors of Cr, Cu, Pb and Zn were 4.6%, 7.5%, 3.8% and 14.2%, respectively, indicating high accuracy. The relative errors of the remaining three elements As, Cd, and Hg are 55.5%, 55.7%, and 37.2%, respectively. The errors are relatively large and will be significantly reduced as the detector accuracy improves in the future. The device can widely be expected to accurately, stably and real-time detect the content of heavy metals in soil. Subsequently, the data can be summarized to the cloud platform, indicating an excellent real-time performance and data integration.