The invention relates to the technical field of nuclear physical detection, in particular to a nuclear latex
proton track positioning and predicting
system based on a bidirectional neural network, and the
system comprises a bidirectional neural track extraction module which is used for carrying out space-time bidirectional
feature coding and capturing a track motion dependency relationship; the multi-scale correlation positioning engine is used for fusing
pyramid pooling and improving YOLOv8 to accurately position track end points and inflection points; the
dynamic noise suppression network is used for generating a GAN construction
noise library, adaptively denoising, retaining track edge details and improving the background
signal-to-
noise ratio; the track evolution prediction module is used for predicting a
proton motion track, constructing a
dynamic prediction model by fusing a graph neural network, controlling a prediction error, and correcting parameters and adjusting a path through actually measured data when the error exceeds a threshold value; and the intelligent scanning control module is used for dynamically optimizing
microscope parameters and paths according to track positioning and prediction results. Therefore, the problems of insufficient positioning precision, poor prediction stability, poor dynamic adaptability and the like in the prior art are solved.