A method for early prediction of wheat yield based on artificial intelligence

The invention is based on the digitization and acceleration of the existing traditional method of farmers for estimating wheat yield by means of artificial intelligence algorithms. In addition to the algorithms, the invention also includes a new base for estimating yields that does not depend on weather conditions, soil type and wheat crop type.

 

The invention proposes a process which includes: detection of wheat ears, automatic counting of ears, and formation of a database for yield prediction in a new way and yield prediction which is done with new data.

 

The detection of wheat ears is based on image processing algorithms and artificial intelligence, ie. deep learning using convolutional neural networks. Ear detection takes place as a scalable phase applicable to both RGB and thermal images of wheat ears.
Automatic class counting is based on digital image processing algorithms.

 

The formation of the database for yield prediction consists of reference data of biomass and number of spikes, then the number of spikes obtained by automatic counting of spikes and new estimated biomass obtained by prediction. After forming the data for the prediction database, the prediction phase itself uses the new estimated biomass and the number of spikes to predict the yield. The prediction phase uses machine learning algorithms, and the prediction results are validated on actual harvest results and yield values by agricultural experiments, at the end of the same season.