Abstract: Leaf nitrogen content is one of the most important indicators of rice health. Non-destructive detection of the leaf nitrogen content is crucial to ensure the growth, development, and yield of rice. However, the existing detectors are expensive and large in size. It is a high requirement for the operating environment. This study aims to fully meet the demand for the low-cost acquisition of the field crop nutrition, in order to reduce the application cost for the portability. Multispectral detectors were also designed using spectral features. Therefore, the experimental area was taken as the Precision Agriculture Aviation Research Base of Shenyang Agricultural University in Gengzhuang Town, Haicheng City, Anshan City, Liaoning Province, China. Ocean Optics HR2000+ was used to collect the spectral reflectance of "Shen Nong 9816" rice leaves. The nitrogen concentration of the leaves was measured using indoor experiments. Then, the competitive adaptive weighted sampling (CARS) was used to extract the characteristic bands. The nitrogen characteristic transfer index (NCTI) was also selected to extract the nitrogen spectral features of the rice leaves. According to the optimization and combination of the nitrogen spectral features, a suitable multispectral sensor was selected to design a portable non-destructive detector of the multispectral rice nitrogen. The hardware system of the device consisted of a host and an external leaf clamp. The host included a spectral acquisition, control, and display, as well as an external power module. The external leaf clamp was used to fix the rice leaves. The host interface of the computer operation was designed using the PyQt5 framework in the Python3.11.4 environment. The acquisition and storage of the spectral data were realized for the display of the rice nitrogen. The detection equipment was developed to collect the spectral reflectance of the rice leaves. The characteristic bands with the vegetation index were used as the inputs, while the nitrogen concentration was used as the output. The rice nitrogen concentration models were constructed using partial least squares regression (PLSR), extreme learning machine (ELM), and extreme learning machine using bat optimization algorithm (BA-ELM). Among them, the accuracy of the BA-ELM inversion was the highest with the spectral features, where the characteristic bands and NCTI were the inputs. The determination coefficient of the model on the training set was 0.792, and the root mean square error was 0.423%. While on the test set, the determination coefficient was 0.783, and the root mean square error was 0.423%. The prediction model was imported into the detection equipment. The accuracy and stability of the equipment were verified after prediction. According to the residual device accuracy of the true and predicted values, the maximum absolute residual value was 0.737%. The coefficient of variation was used to verify the stability of the equipment, and the maximum coefficient of variation was 1.901%. The portable detection equipment of the rice nitrogen was independently developed for high accuracy and stability. The finding can fully meet the requirement for the real-time detection of the rice nitrogen concentration.