Variable fertilizing requires the rapid and in-situ high-accuracy collection of farmland soil nutrients information. However, existing equipment could not meet the needs of field measurement in precision agriculture. Hence, a vehicle-mounted in-situ soil parameters detector was developed based on near-infrared (NIR) diffuse reflection. The detector used a tungsten halogen light source with better illumination stability instead of sunlight to perform soil spectrum detection to improve the adaptability of the instrument to working conditions. A soil total nitrogen measurement extreme learning machine model consisting of seven sensitive wavelengths (1 070, 1 130, 1 245, 1 375, 1 450, 1 550, 1 680 nm) was developed to improve the real-time measurement accuracy. The detector consisted of a mechanical part, optical part and control part. The mechanical part provided platform support for the detector, The optical part was composed of a halogen tungsten light source, a light source adapter flange, a NIR guiding fiber, and a set of detection assembly including an incident light exit end, seven InGaAS photodetectors, and seven single-band filters. The control part realized the collection and processing of the soil measurement signal with a MSP430F149 main control chip module When the detector performed farmland soil nutrients detection, and the tungsten halogen light source transmitted the detection light source to the surface of the detection soil through the NIR guiding fiber and incident light exit end of the detection assembly. The diffuse light from the surface of the detection soil was used to detect soil nutrient parameters. A light source adapter flange at the junction of the tungsten halogen light source and the NIR guiding fiber was designed to minimize the loss of the detection source during transmission. The filter of detection assembly filtered the diffuse light to become a single-band detection light, and the InGaAS photodetector realized photoelectric conversion of the single-band detection light, and the signal processing unit calculated the reflectance at each sensitive wavelength. After the development of the detector was completed, a standard gray board was used as the measurement object to conduct the optical calibration test. The test results showed that the correlation coefficient (R) between the reflectance value of the detector at seven sensitive wavelengths and the reflectance value of the MATRIX-I type Fourier spectrum analyzer had a maximum of 0.997 8 and an average of 0.927 8, which indicated that the detector had higher detection accuracy. In order to further evaluate the detection accuracy of farmland nutrients content using the detector, and the farmland application test of the detector was carried out at the Tongzhou Experimental Station of China Agricultural University. The test results showed that the correlation coefficient (R) between the measured value of soil nutrients content using the detector and the laboratory standard test method were all above 0.90. The test results showed that the vehicle-mounted in-situ soil parameters detector could realize rapid in-situ high-accuracy collection of farmland nutrients information.