Linear ground feature repetition type classification method based on spatial relation coding

The invention discloses a linear ground feature repetition type classification method based on spatial relation coding, which comprises the following steps: identifying bus route repetition types, classifying the route repetition types, obtaining various spatial relation types through abstraction so as to determine the spatial relation of bus routes, and then combining with a corresponding feature extraction mode so as to determine the spatial relation of the bus routes. The method comprises the following steps: constructing a deep learning model, completing identification of different relation types by adopting a loss calculation method, constructing the deep learning model, designing to extract trajectory features through a bidirectional LSTM, and adding two full connection layers Lear1 and Lear2 to further process the trajectory features extracted by the BiLSTM so as to obtain more suitable vector representation. Through loss calculation, optimization of the weight and model training of the whole classific