Retrieval of horticultural crop morphology from color based on Elman neural network

The quantification of the relationship between morphological and color indicators in various organs of horticultural crops is of great significance for crop digital visualization research using computer vision technology. To study this relationship, observational data from a six-year experiment were collected, focusing on seven kinds of color component values of different organs including root, stem, and leaf. Using the collected color data as input, a simulation model was established based on the Elman neural network for six horticultural crops including zizania, cucumber, celery, spinach, parsley, and tea. Results indicated that the horticultural crop morphology model based on the Elman neural network exhibited high simulation accuracy with root mean square error (RMSE) ranging from 0.14 to 1.05 cm and normalized root mean square error (NRMSE) ranging from 2.02% to 11.34% for the maximum root length simulation model. The simulation model for stem length and diameter had an RMSE ranging from 1.42 to 4.96 cm and 0.25 to 1.17 mm, respectively, with NRMSE ranging from 18.19% to 25.65% and 15.13% to 27.25%, respectively. Similarly, chlorophyll content, leaf length, leaf width, and leaf area simulation models exhibited RMSE ranging from 2.80 to 8.22 SPAD, 0.44 to 18.04 cm, 0.22 to 3.49 cm, and 0.25 to 36.39 cm², respectively, with NRMSE ranging from 8.63% to 21.04%, 15.00% to 22.87%, 15.12% to 33.58%, and 6.88% to 24.90%, respectively. These findings provide essential theoretical support for precision agriculture in areas of water and fertilizer management, plant growth diagnosis, and yield prediction.