The invention discloses a multi-
modal data fusion method, device, equipment and medium, and relates to the technical field of nuclear biochemical situation analysis, and the method comprises the steps: carrying out the
imaging processing of structured
detector data through a GNN model, generating input image data, carrying out the regional
pollutant concentration distribution prediction of the image data through a ResNet-LSTM model, and carrying out the prediction of the regional
pollutant concentration distribution through the ResNet-LSTM model. The ResNet-LSTM network uses a multilayer convolutional network and a residual module to extract
spatial distribution characteristics in the image data, uses a long-
short term memory network (LSTM) and an attention layer to extract
time distribution characteristics in the two-dimensional image data, and performs fusion application of two loss functions of a deviation rate and a structural coefficient to obtain a two-dimensional image. And the space-time prediction capability of the model on the nuclear biochemical
pollutant distribution situation is fully exerted. According to the method, effective fusion of structured detection data and image data is realized through a
deep learning technology, and the situation analysis capability of nuclear and
biochemical detection results is remarkably improved.