IntegrateNet: A Deep Learning Network for Maize Stand Counting From UAV Imagery by Integrating Density and Local Count Maps
Wenxin Liu, Jing Zhou, Biwen Wang, Martin Costa, Shawn M. Kaeppler, Zhou Zhang
Abstract
Crop stand count plays an important role in modern agriculture as a reference for precision management and plant breeding. In this study, a new network—IntegrateNet—was proposed to supervise the learning of density map and local count simultaneously and thus boost the model performance by balancing the tradeoff between their errors. The IntegrateNet was trained and validated with an image set containing 124 maize images by an unmanned aerial vehicle. The model achieved an excellent result for 24 test images with the root-mean-square error of 2.28 and the coefficient of determination (<inline-formula> <tex-math notation="LaTeX">$R^{2}$ </tex-math></inline-formula>) of 0.9578 between the predicted and ground-truth maize stand counts. In conclusion, the proposed model provides an efficient solution for counting maize stands at early stages and could be used as a reference for similar studies.