Abstract: Abstract: Huanglongbing (HLB) is known as one of the most important diseases in citrus production. Early detection of diseased plants can be used to prevent the disease from the spreading, in order to reduce the severity of the disease. Hyperspectral analysis can be expected in the crop disease detection in recent years, due to the rich spectral information. However, the hyperspectral equipment is too expensive to apply in large scale, particularly with a large amount of calculation for a large number of bands. An accurate band selection can be utilized to remove the redundant information for the less data storage, calculation, and transmission, in order to avoid the "dimension disaster", and equipment costs. In this study, an Exemplar Component Analysis (ECA)-based method was proposed to optimize the characteristic wavebands of HLB diseases using a ground feature spectrometer. The spectrum detector was also selected the spectral characteristic bands from the hyperspectral information of citrus plants. Three algorithms were utilized to optimize the combination of sever spectral wavebands for comparison. The optimal bands were obtained to evaluate the robustness of the four band selection. Among them, the six machine learning methods were used for the modeling and classification. In addition, a multi-spectrometer was designed using the optimal characteristic band for the detection of citrus HLB. The results show that an accuracy of more than 92% was achieved in the test set of the feature band that selected by the ECA algorithm combined with the six classifiers, indicating the excellent robustness. The self-developed multi-spectrometer was detected up to 95% accuracy for the HLB using the characteristic band. Experiments show that the multiple spectrometer was feasible to characterize the HLB for a small number of characteristic bands. A reasonable characteristic band was greatly reduced the design cost of spectrum detection in the special agricultural disease, and then improved the accuracy of disease prevention and control in orchards. As such, the HLB spectral feature bands were extracted to develop a spectral detector. The proposed ECA-based band selection can be expected to obtain the characteristic bands of citrus HLB with the high robustness. Specifically, the healthy leaves were distinguished from the HLB leaves, particularly from the high similarity to the HLB disease due to deficiency. Moreover, the ECA-based band preference was only preferred seven bands. The redundant bands were removed to reduce the data transmission and storage in the application scenario of large number of samples. The comparison of different classifiers showed that the Xgboost performed the best, in terms of the stable spectral data. The high detection accuracy was also achieved in the rest classifiers using the feature bands. A low-cost multispectral instrument was also designed to promote the unstable performance in different band selection. Anyway, the accurate HLB detection was obtained using the spectrometer data with the feature bands model, where the highest recognition accuracy was close to 95% in the case of triple classification. The promising application prospect was given in the future, compared with the rest high precision and high price professional instruments. The data analysis and spectrometer can be affordable to promote the smart agriculture, compared with the generally expensive spectrometers on sale. Low-cost spectrometers are more likely to be popularized, in order to reduce the amount of generated data for the less hardware requirements. A balance between detection accuracy and production cost can lay a strong foundation for the wide popularization of intelligent equipment.