Pesticides play an important role in increasing agricultural production, but their growing use has led to pesticide residues accumulating in soil and seeping into groundwater, causing water quality pollution and further affecting ecosystems and human health. Currently, traditional groundwater pesticide detection methods such as gas chromatography and liquid chromatography are technically mature but suffer from drawbacks like high cost, complex operation, and low detection frequency. In contrast, in-situ monitoring avoids pollutant loss during sample collection and handling and enables real-time, on-site detection. Fluorescence spectroscopy, known for its high sensitivity and real-time capabilities, holds great potential for detecting organic pollutants in groundwater. Among these, three-dimensional fluorescence spectroscopy can identify and quantify pesticides even in complex background interference conditions. However, due to the narrow space of groundwater monitoring wells, typically only 5~10 cm in diameter, large-scale instruments are impractical for in-situ use. Moreover, conventional single-wavelength excitation and detection methods lack selectivity. As a result, technical and equipment limitations remain major challenges for in-situ fluorescence detection in this field.In this study, a novel multi-channel in-situ fluorescence rapid detection system was designed for the on-site detection of pesticide pollutants in groundwater. The system, based on a deep ultraviolet light-emitting diode and photomultiplier tube, enables non-intrusive real-time acquisition of fluorescence signals. A Support Vector Machine (SVM) was used for pesticide component identification, and Multiple Linear Regression (MLR) was applied for pesticide concentration inversion. The system demonstrated good responsiveness to pesticides. In deionized water, pesticide fluorescence intensity showed a strong linear correlation with concentration, with fitting coefficients of 0.990 0 or higher. The Limits Of Detection (LOD) for chlorothalonil in the 255~340 nm and 265~340 nm channels are 0.1 μg·L?1and 0.3 μg·L?1, respectively; for chlorpyrifos in the 265~340 nm and 285~340 nm channels, both are 6.5 μg·L?1; for diniconazole in the 255~340 nm, 265~340 nm, and 285~340 nm channels, the LODs are 0.002, 0.003, and 0.002 μg·L?1, respectively; for carbaryl, the LODs are 0.05, 0.04, and 0.08 μg·L?1; and for carbendazim, the LODs are 2.0, 1.7, and 1.7 μg·L???1, respectively. In groundwater samples, the linear fitting coefficients between fluorescence intensity and pesticide concentration were above 0.958 9. The lowest LODs for each pesticide are 0.4 μg·L?1for chlorothalonil (265~340 nm), 6.9 μg·L?1 for chlorpyrifos (285~340 nm), 0.002 μg·L?1 for diniconazole (285~340 nm), 0.06 μg·L?1 for carbaryl (265~340 nm), and 1.9 μg·L?1 for carbendazim (285~340 nm). Using the SVM and MLR algorithms to analyze the fluorescence data obtained from groundwater pesticide samples, the system achieved selective and quantitative detection of multiple pesticides. The accuracy for single-component samples is 92.5%, and for mixed samples, 94.1%. The average relative error in quantitative analysis is 0.8% for single-component samples and 8.3% for mixed samples, indicating high precision in both identification and quantification. The results demonstrate that the in-situ multi-channel fluorescence detection system, combined with machine learning algorithms, enables selective and quantitative detection of pesticide pollutants in groundwater.