Three-dimensional radar recessive disease identification method based on feature pyramid network

The invention discloses a three-dimensional radar recessive disease identification method based on a feature pyramid network, and the method comprises the steps: firstly guaranteeing the consistency and availability of three-dimensional radar data through the preprocessing and quality improvement of low-quality, non-cooperative and multi-source heterogeneous data, including data cleaning, noise reduction, feature selection, data standardization and other methods; a multi-modal data fusion technology is introduced and is used for integrating various data sources and enhancing the information amount of the data; secondly, designing and constructing the application of a feature pyramid network architecture on three-dimensional radar data through a Faster-R-CNN-FPN network, designing a loss function, adopting a training strategy, enabling targets of different sizes to have proper feature representation under corresponding scales, fusing multi-scale information, predicting the targets of different sizes under diff