This study aims to design a plant pigment measurement system using digital images and deep learning, incorporating various color spaces including RGB, HSV, LAB, and YCbCr. The proposed method serves as a faster, more cost-effective, and accurate alternative to traditional methods such as spectrophotometric analysis and HPLC. Experimental results indicate that the choice of color space and inpaint preprocessing settings significantly impacts the accuracy of the CNN P3Net model. The combination of RGB+YCbCr with inpaint and RGB+LAB without inpaint yielded the lowest validation MAE values. The study also demonstrates that color constancy phenomena influence model accuracy, with color spaces that account for this phenomenon, such as RGB+YCbCr with inpaint, providing better accuracy than those that do not.