The invention discloses a self-adaptive
feature fusion genotype-
phenotype prediction method for forest intelligent breeding, and relates to the technical field of forest breeding and intelligent
crossover, and the method comprises the specific steps of multi-
modal data acquisition, data preprocessing, self-adaptive
feature fusion,
phenotype prediction model construction and
phenotype prediction. According to the method, self-adaptive fusion of multi-
modal data is realized through
principal component analysis and a feature correlation enhancement technology, the problem of high-dimensional
data redundancy is solved, core correlation information of each
modal data is reserved, SNP sites and environmental factors are screened through a mixed
linear model, a Transform
encoder is embedded on the basis of a multi-layer
perceptron, and the accuracy and the reliability of the
system are improved. The multi-task deep neural network is used for capturing the nonlinear interaction effect of the
genotype and the environment factor, the complex association among the
genotype, the environment and the phenotype data can be described more accurately, meanwhile, the multi-task deep neural network dynamically balances the loss contribution of different phenotype tasks through a multi-phenotype task adaptive
weight distribution formula, and the accuracy of the multi-task deep neural network is improved. Cooperative prediction of multiple characters such as tree height,
diameter at breast height and crown breadth is realized.