The invention provides a
deep learning-based multi-
modal phenotype determination method for economic traits of
lateolabrax japonicus, and relates to the technical field of high-
throughput phenotype determination of economic traits of
lateolabrax japonicus. According to the
deep learning-based multi-
modal phenotype determination method, multi-angle video streams are synchronously acquired by utilizing a
sensor array, a temporary visual identifier is allocated to each
tail lateolabrax japonicus, and then, a visual identifier is allocated to each
tail lateolabrax japonicus; when a preset
underwater measurement condition is met, a video clip in a time window is called, the
divergence value of a multi-view
motion vector field is calculated, a group of optimal key frames are selected from the clip in combination with the body principal axis direction and the
lateral line contour definition, multi-view
feature fusion and deformation compensation based on the
divergence value are performed on the optimal
key frame group, and the multi-view
motion vector field is obtained. The method comprises the following steps: extracting static three-dimensional phenotypic characteristics, calculating dynamic behavior characteristics from a motion track, collecting the two types of characteristics according to identifiers to form
underwater multi-
modal phenotypic vectors, inputting the multi-modal phenotypic vector set of a group into a pre-trained economic character prediction model, and outputting individual predictive character indexes and group statistical results in batches.