Diamond, an important electronic and optoelectronic material, exhibits excellent performance in deep ultraviolet photodetection. However, its lack of visible light response and nonvolatile photocurrent hinders its application in neuromorphic sensors that mimic the human eye. In this work, we report a diamond-based neuromorphic retina perception system (NRPS) that demonstrates broad-spectrum photoresponse beyond the human visual range, nonvolatile photocurrent, and wavelength sensitivity. This outcome is achieved by introducing a large number of defects and doping states into the diamond film through heteroepitaxial growth on the SiO2 substrate and Si doping. By adjusting the input light and electrical stimulation, human eye's pupil and image memory behavior are successfully simulated. More interestingly, the diamond NRPS exhibits a significant difference in the decay characteristic of photocurrent under different wavelengths of illumination. Based on this characteristic, the NRPS achieves recognition of unknown colors and trajectory tracking of dynamic light dots. Furthermore, the image recognition accuracy is improved from 67.5% to 81.7% by filtering the visible light information and extracting useful ultraviolet light information with the device. This study offers an effective strategy to develop a diamond-based neuromorphic sensor.