A rapid and accurate diagnosis of nutritional status in field crops is crucial for site-specific fertilizer management. The micro near-infrared spectrometer (Micro-NIRS) is an extremely portable optical device that can be connected to a smartphone through a Bluetooth connection. In this study, a Micro-NIRS was used to evaluate pigment contents, namely chlorophyll a (Chl-a), chlorophyll b (Chl-b), and carotenoid (Car) in two varieties of field tea plants. A variable combination population analysis (VCPA), genetic algorithm (GA), and VCPA-GA hybrid strategy were used to select characteristic wavelengths; a partial least squares regression (PLSR) algorithm was employed for modeling. Results indicated that the simplified VCPA-GA-PLSR models provided the most favorable performance among all models for Chl-a, Chl-b, and Car content prediction; the correlation coefficients in prediction (Rps) were 0.9226, 0.9006, and 0.8313, respectively; the root mean square errors in prediction (RMSEPs) were 0.0952, 0.0771, and 0.0373 mg/g, respectively; the relative prediction deviations (RPDs) were 2.55, 1.92, and 1.79, respectively. Extracted characteristic variables occupied <13.63% of full spectra. The current work provided a useful example for implementing a smartphone-based Micro-NIRS system that can diagnose plant nutrition rapidly, nondestructively, and at low cost.