Combined with photoelectric detection technology and UV-Vis absorption spectroscopy in spectral analysis technology, a multi-parameter portable surface water quality detection system is developed, which can quickly detect phosphate, nitrite, and chemical oxygen demand (COD) and ammonia nitrogen water quality parameters on site. For the substances in the water body that absorb the characteristic wavelengths in the visible part, a camera is used to collect the visible spectrum, and the grayscale image of the visible spectrum image is modeled by a convolutional neural network. The concentration value of substances whose absorption characteristic wavelength is in the ultraviolet band is measured by photoelectric detection technology. The established convolutional neural network model is transplanted into ZYNQ, combined with ultraviolet photoelectric sensor, the concentration value of the detected substance is displayed on the LCD, so as to realize the economy, portability, real-time and rapidity of the water quality detector. Research indicates: The prediction value of convolutional neural network is obtained as the tendency value of sample solution in 8 output types of concentration value, the highest accuracy is 100%, and the lowest is 40%. The highest error of COD concentration value is 10%, which proves that the detection system has practical value with low noise and high precision.