The invention discloses a training method of agricultural directional
artificial intelligence. Graphic data of
agricultural land and corresponding agricultural data are continuously collected, and pretreatment is performed, a
training set and a
test set are performed, a
simulation platform is built, a corresponding
artificial intelligence algorithm is selected according to the performance of thecomputing terminal, training is performed through a parallel virtual
system; calculation power distribution is performed according to performance requirements in various
artificial intelligence algorithms, an evaluation network is constructed, each decision is evaluated, the decision training efficiency is improved, the decision made by each grid is planned as a whole, the cost of the made corresponding decision is selected according to the grid distance, and the obtained
optimal decision is deployed. According to the method, training is carried out in a single operation node through multipleartificial intelligence
training methods, more parallel operations of a CPU and a GPU are mobilized under the condition that the computing power is limited, an
evaluation function is introduced to reduce the training expenditure, and the operation cost of agricultural execution is greatly reduced.