Click on “Download PDF” for the PDF version or on the title for the HTML version.

If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.

Abstract. Crop breeding is a promising solution for the food crisis in 2050 by developing new crop varieties with improved traits, including high yield potential. In breeding programs, yield is always used as a primary trait to select superior genotypes and evaluate breeding efficiency. Yield trials have to be conducted in multiple locations in several years to evaluate the performance of numerous breeding lines and their adaptability to environments, and yield data are collected manually in these trials. To improve the efficiency of yield trials in soybean breeding programs, a yield estimation method using Unmanned Aerial Vehicle imagery and convolutional neuron network (CNN) was proposed in this paper. A group of 972 breeding lines in three maturity groups was planted under rainfed conditions for a soybean drought tolerance study. Aerial images were taken at the V6, R1 and R6 growth stages. Seven image features representing plant height, canopy color and canopy texture were extracted. A mixed CNN model was built to estimate soybean yield by taking the multi-layer 2D image features and two categorical factors, i.e. maturity groups and drought tolerance, as predictors. Results show that the correlation (R²) between the predicted yield and the measured yield is up to 78% with the root mean square error of 391 kg·ha^-1 (33.8% of the average yield), leading to the conclusion that the UAV imagery and deep learning methods are promising in estimating yield for soybean breeding purposes.