The Technology
Ear density is a key wheat yield determinant but manual counting is time-consuming, subjective and non-standardised. Automatic ear counting systems have been developed using image -based crop recognition combined with machine learning. Technology already developed for handheld RGB and more reliable thermal camera systems, could be readily adapted for drone use. Technologies are already enabled to compensate for different light intensities and soil reflectance. Open-source datasets used to instruct machine learning for UK crops are available from Nottingham University researchers, which facilitate recognition at accuracies of 95.9% for spikes and 99.7% for spikelets.
How it Works
Thermally or conventionally captured RGB images are analysed by deep learning algorithms which have been taught how to recognise wheat ears using convolutional neural networks (CNN). These neural networks transform individual images by putting them through a series of hidden layers made up of a set of neurons, organised in 3 dimensions i.e. width, height and depth. The neurons on one layer only connect to a small region of those in the next layer. This process detects wheat ear features. The final image is reduced to a single vector aligned along the depth dimension, which has been assigned the highest probability for the image object being the wheat ear that the algorithm predicts it should be. The method is well known for face recognition software.
What are the Benefits to a Farmer / Agronomist
Fast evaluation of ear density helps monitor the efficiency of crop management practices, gives an early prediction of grain yield and is ideal for phenotyping cultivar traits in breeding programmes.
Key researchers/stakeholders
Nottingham University
Barcelona University
01945 461177
Email Us