Abstract:Moisture content is a key indicator reflecting the physical and chemical properties and quality control of substances, which is particularly important in many fields such as industry and agriculture. Aiming at the problems of high power consumption, complex light source structure and limited measurement accuracy of traditional near-infrared moisture detection equipment, this paper designs and implements a near-infrared moisture detection system based on dual-band LED and optical convergence structure. The system uses a 1450 nm and 1300 nm dual-band LED array in a spherical alternating layout, and combines a plano-convex lens to converge the light path, which improves the utilization efficiency of diffuse light and the uniformity of the light field. In terms of signal processing, this paper establishes the denoising mechanism of " hardware median filtering + wavelet transform ", introduces relative absorbance and its derivative features, and constructs PLS-SVR series model to predict moisture content. In this study, sand and wheat were selected as verification objects. The experimental results showed that the determination coefficient ( R2 ) of the test set of sand and stone samples reached 0.9624, the root mean square error of prediction ( RMSEP ) was 0.4277 %, and the residual predictive deviation ( RPD ) was 5.16. The determination coefficient ( R2 ) of the test set of wheat samples reached 0.9201, the root mean square error of prediction ( RMSEP ) was 1.3813 %, and the residual predictive deviation ( RPD ) was 3.54. The results show that the system has the characteristics of simple structure, stable light source and high detection accuracy, which can provide reference for miniaturized moisture detection equipment based on semiconductor light source.