Sea radar video target detection method based on fully convolutional single-stage network

The invention relates to a sea radar video target detection method based on a fully convolutional single-stage network. The method provided by the invention comprises the steps of performing preprocessing and signal processing on an original echo of a sea exploration radar; constructing a sea radar target detection data set by referring to multi-frame radar video data and corresponding radar target constant false alarm detection trace data; resNet34 is used as a trunk feature extractor, a full-convolution single-stage detection network is constructed, and the full-convolution single-stage detection network is adopted to train radar target detection samples in a training set; testing the full-convolution single-stage detector obtained by training by using a test set sample, and recording a test index; and when the iteration reaches the maximum value, outputting an output full-convolution single-stage detector with the optimal performance index in all iterations according to the recorded test indexes, and input